Search Results

Search found 41110 results on 1645 pages for 'oracle integration solution'.

Page 566/1645 | < Previous Page | 562 563 564 565 566 567 568 569 570 571 572 573  | Next Page >

• Verizon Wireless Supports its Mission-Critical Employee Portal with MySQL

  - by Bertrand Matthelié

  Normal 0 false false false EN-US X-NONE X-NONE /* Style Definitions */ table.MsoNormalTable {mso-style-name:"Table Normal"; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-style-noshow:yes; mso-style-priority:99; mso-style-qformat:yes; mso-style-parent:""; mso-padding-alt:0cm 5.4pt 0cm 5.4pt; mso-para-margin:0cm; mso-para-margin-bottom:.0001pt; mso-pagination:widow-orphan; font-size:11.0pt; font-family:"Cambria","serif"; mso-ascii-font-family:Cambria; mso-ascii-theme-font:minor-latin; mso-fareast-font-family:"Times New Roman"; mso-fareast-theme-font:minor-fareast; mso-hansi-font-family:Cambria; mso-hansi-theme-font:minor-latin;} Verizon Wireless, the #1 mobile carrier in the United States, operates the nation’s largest 3G and 4G LTE network, with the most subscribers (109 millions) and the highest revenue ($70.2 Billion in 2011). Verizon Wireless built the first wide-area wireless broadband network and delivered the first wireless consumer 3G multimedia service in the US, and offers global voice and data services in more than 200 destinations around the world. To support 4.2 million daily wireless transactions and 493,000 calls and emails transactions produced by 94.2 million retail customers, Verizon Wireless employs over 78,000 employees with area headquarters across the United States. The Business Challenge Seeing the stupendous rise in social media, video streaming, live broadcasting…etc which redefined the scope of technology, Verizon Wireless, as a technology savvy company, wanted to provide a platform to its employees where they could network socially, view and host microsites, stream live videos, blog and provide the latest news. The IT team at Verizon Wireless had abundant experience with various technology platforms to support the huge number of applications in the company. However, open-source products weren’t yet widely used in the organization and the team had the ambition to adopt such technologies and see if the architecture could meet Verizon Wireless’ rigid requirements. After evaluating a few solutions, the IT team decided to use the LAMP stack for Vzweb, its mission-critical, 24x7 employee portal, with Drupal as the front end and MySQL on Linux as the backend, and for a few other internal websites also on MySQL. The MySQL Solution Verizon Wireless started to support its employee portal, Vzweb, its online streaming website, Vztube, and internal wiki pages, Vzwiki, with MySQL 5.1 in 2010. Vzweb is the main internal communication channel for Verizon Wireless, while Vztube hosts important company-wide webcasts regularly for executive-level announcements, so both channels have to be live and accessible all the time for its 78,000 employees across the United States. However during the initial deployment of the MySQL based Intranet, the application experienced performance issues. High connection spikes occurred causing slow user response time, and the IT team applied workarounds to continue the service. A number of key performance indexes (KPI) for the infrastructure were identified and the operational framework redesigned to support a more robust website and conform to the 99.985% uptime SLA (Service-Level Agreement). The MySQL DBA team made a series of upgrades in MySQL: Step 1: Moved from MyISAM to InnoDB storage engine in 2010 Step 2: Upgraded to the latest MySQL 5.1.54 release in 2010 Step 3: Upgraded from MySQL 5.1 to the latest GA release MySQL 5.5 in 2011, and leveraging MySQL Thread Pool as part of MySQL Enterprise Edition to scale better After making those changes, the team saw a much better response time during high concurrency use cases, and achieved an amazing performance improvement of 1400%! In January 2011, Verizon CEO, Ivan Seidenberg, announced the iPhone launch during the opening keynote at Consumer Electronic Show (CES) in Las Vegas, and that presentation was streamed live to its 78,000 employees. The event was broadcasted flawlessly with MySQL as the database. Later in 2011, Hurricane Irene attacked the East Coast of United States and caused major life and financial damages. During the hurricane, the team directed more traffic to its west coast data center to avoid potential infrastructure damage in the East Coast. Such transition was executed smoothly and even though the geographical distance became longer for the East Coast users, there was no impact in the performance of Vzweb and Vztube, and the SLA goal was achieved. “MySQL is the key component of Verizon Wireless’ mission-critical employee portal application,” said Shivinder Singh, senior DBA at Verizon Wireless. “We achieved 1400% performance improvement by moving from the MyISAM storage engine to InnoDB, upgrading to the latest GA release MySQL 5.5, and using the MySQL Thread Pool to support high concurrent user connections. MySQL has become part of our IT infrastructure, on which potentially more future applications will be built.” To learn more about MySQL Enterprise Edition, Get our Product Guide.

  Read the article

• Java EE 6 and NoSQL/MongoDB on GlassFish using JPA and EclipseLink 2.4 (TOTD #175)

  - by arungupta

  TOTD #166 explained how to use MongoDB in your Java EE 6 applications. The code in that tip used the APIs exposed by the MongoDB Java driver and so requires you to learn a new API. However if you are building Java EE 6 applications then you are already familiar with Java Persistence API (JPA). Eclipse Link 2.4, scheduled to release as part of Eclipse Juno, provides support for NoSQL databases by mapping a JPA entity to a document. Their wiki provides complete explanation of how the mapping is done. This Tip Of The Day (TOTD) will show how you can leverage that support in your Java EE 6 applications deployed on GlassFish 3.1.2. Before we dig into the code, here are the key concepts ... A POJO is mapped to a NoSQL data source using @NoSQL or <no-sql> element in "persistence.xml". A subset of JPQL and Criteria query are supported, based upon the underlying data store Connection properties are defined in "persistence.xml" Now, lets lets take a look at the code ... Download the latest EclipseLink 2.4 Nightly Bundle. There is a Installer, Source, and Bundle - make sure to download the Bundle link (20120410) and unzip. Download GlassFish 3.1.2 zip and unzip. Install the Eclipse Link 2.4 JARs in GlassFish Remove the following JARs from "glassfish/modules": org.eclipse.persistence.antlr.jar org.eclipse.persistence.asm.jar org.eclipse.persistence.core.jar org.eclipse.persistence.jpa.jar org.eclipse.persistence.jpa.modelgen.jar org.eclipse.persistence.moxy.jar org.eclipse.persistence.oracle.jar Add the following JARs from Eclipse Link 2.4 nightly build to "glassfish/modules": org.eclipse.persistence.antlr_3.2.0.v201107111232.jar org.eclipse.persistence.asm_3.3.1.v201107111215.jar org.eclipse.persistence.core.jpql_2.4.0.v20120407-r11132.jar org.eclipse.persistence.core_2.4.0.v20120407-r11132.jar org.eclipse.persistence.jpa.jpql_2.0.0.v20120407-r11132.jar org.eclipse.persistence.jpa.modelgen_2.4.0.v20120407-r11132.jar org.eclipse.persistence.jpa_2.4.0.v20120407-r11132.jar org.eclipse.persistence.moxy_2.4.0.v20120407-r11132.jar org.eclipse.persistence.nosql_2.4.0.v20120407-r11132.jar org.eclipse.persistence.oracle_2.4.0.v20120407-r11132.jar Start MongoDB Download latest MongoDB from here (2.0.4 as of this writing). Create the default data directory for MongoDB as: sudo mkdir -p /data/db/sudo chown `id -u` /data/db Refer to Quickstart for more details. Start MongoDB as: arungup-mac:mongodb-osx-x86_64-2.0.4 <arungup> ->./bin/mongod./bin/mongod --help for help and startup optionsMon Apr  9 12:56:02 [initandlisten] MongoDB starting : pid=3124 port=27017 dbpath=/data/db/ 64-bit host=arungup-mac.localMon Apr  9 12:56:02 [initandlisten] db version v2.0.4, pdfile version 4.5Mon Apr  9 12:56:02 [initandlisten] git version: 329f3c47fe8136c03392c8f0e548506cb21f8ebfMon Apr  9 12:56:02 [initandlisten] build info: Darwin erh2.10gen.cc 9.8.0 Darwin Kernel Version 9.8.0: Wed Jul 15 16:55:01 PDT 2009; root:xnu-1228.15.4~1/RELEASE_I386 i386 BOOST_LIB_VERSION=1_40Mon Apr  9 12:56:02 [initandlisten] options: {}Mon Apr  9 12:56:02 [initandlisten] journal dir=/data/db/journalMon Apr  9 12:56:02 [initandlisten] recover : no journal files present, no recovery neededMon Apr  9 12:56:02 [websvr] admin web console waiting for connections on port 28017Mon Apr  9 12:56:02 [initandlisten] waiting for connections on port 27017 Check out the JPA/NoSQL sample from SVN repository. The complete source code built in this TOTD can be downloaded here. Create Java EE 6 web app Create a Java EE 6 Maven web app as: mvn archetype:generate -DarchetypeGroupId=org.codehaus.mojo.archetypes -DarchetypeArtifactId=webapp-javaee6 -DgroupId=model -DartifactId=javaee-nosql -DarchetypeVersion=1.5 -DinteractiveMode=false Copy the model files from the checked out workspace to the generated project as: cd javaee-nosqlcp -r ~/code/workspaces/org.eclipse.persistence.example.jpa.nosql.mongo/src/model src/main/java Copy "persistence.xml" mkdir src/main/resources cp -r ~/code/workspaces/org.eclipse.persistence.example.jpa.nosql.mongo/src/META-INF ./src/main/resources Add the following dependencies: <dependency> <groupId>org.eclipse.persistence</groupId> <artifactId>org.eclipse.persistence.jpa</artifactId> <version>2.4.0-SNAPSHOT</version> <scope>provided</scope></dependency><dependency> <groupId>org.eclipse.persistence</groupId> <artifactId>org.eclipse.persistence.nosql</artifactId> <version>2.4.0-SNAPSHOT</version></dependency><dependency> <groupId>org.mongodb</groupId> <artifactId>mongo-java-driver</artifactId> <version>2.7.3</version></dependency> The first one is for the EclipseLink latest APIs, the second one is for EclipseLink/NoSQL support, and the last one is the MongoDB Java driver. And the following repository: <repositories> <repository> <id>EclipseLink Repo</id> <url>http://www.eclipse.org/downloads/download.php?r=1&amp;nf=1&amp;file=/rt/eclipselink/maven.repo</url> <snapshots> <enabled>true</enabled> </snapshots> </repository>  </repositories> Copy the "Test.java" to the generated project: mkdir src/main/java/examplecp -r ~/code/workspaces/org.eclipse.persistence.example.jpa.nosql.mongo/src/example/Test.java ./src/main/java/example/ This file contains the source code to CRUD the JPA entity to MongoDB. This sample is explained in detail on EclipseLink wiki. Create a new Servlet in "example" directory as: package example;import java.io.IOException;import java.io.PrintWriter;import javax.servlet.ServletException;import javax.servlet.annotation.WebServlet;import javax.servlet.http.HttpServlet;import javax.servlet.http.HttpServletRequest;import javax.servlet.http.HttpServletResponse;/** * @author Arun Gupta */@WebServlet(name = "TestServlet", urlPatterns = {"/TestServlet"})public class TestServlet extends HttpServlet { protected void processRequest(HttpServletRequest request, HttpServletResponse response) throws ServletException, IOException { response.setContentType("text/html;charset=UTF-8"); PrintWriter out = response.getWriter(); try { out.println("<html>"); out.println("<head>"); out.println("<title>Servlet TestServlet</title>"); out.println("</head>"); out.println("<body>"); out.println("<h1>Servlet TestServlet at " + request.getContextPath() + "</h1>"); try { Test.main(null); } catch (Exception ex) { ex.printStackTrace(); } out.println("</body>"); out.println("</html>"); } finally { out.close(); } } @Override protected void doGet(HttpServletRequest request, HttpServletResponse response) throws ServletException, IOException { processRequest(request, response); } @Override protected void doPost(HttpServletRequest request, HttpServletResponse response) throws ServletException, IOException { processRequest(request, response); }} Build the project and deploy it as: mvn clean packageglassfish3/bin/asadmin deploy --force=true target/javaee-nosql-1.0-SNAPSHOT.war Accessing http://localhost:8080/javaee-nosql/TestServlet shows the following messages in the server.log: connecting(EISLogin( platform=> MongoPlatform user name=> "" MongoConnectionSpec())) . . .Connected: User: Database: 2.7  Version: 2.7 . . .Executing MappedInteraction() spec => null properties => {mongo.collection=CUSTOMER, mongo.operation=INSERT} input => [DatabaseRecord( CUSTOMER._id => 4F848E2BDA0670307E2A8FA4 CUSTOMER.NAME => AMCE)]. . .Data access result: [{TOTALCOST=757.0, ORDERLINES=[{DESCRIPTION=table, LINENUMBER=1, COST=300.0}, {DESCRIPTION=balls, LINENUMBER=2, COST=5.0}, {DESCRIPTION=rackets, LINENUMBER=3, COST=15.0}, {DESCRIPTION=net, LINENUMBER=4, COST=2.0}, {DESCRIPTION=shipping, LINENUMBER=5, COST=80.0}, {DESCRIPTION=handling, LINENUMBER=6, COST=55.0},{DESCRIPTION=tax, LINENUMBER=7, COST=300.0}], SHIPPINGADDRESS=[{POSTALCODE=L5J1H7, PROVINCE=ON, COUNTRY=Canada, CITY=Ottawa,STREET=17 Jane St.}], VERSION=2, _id=4F848E2BDA0670307E2A8FA8,DESCRIPTION=Pingpong table, CUSTOMER__id=4F848E2BDA0670307E2A8FA7, BILLINGADDRESS=[{POSTALCODE=L5J1H8, PROVINCE=ON, COUNTRY=Canada, CITY=Ottawa, STREET=7 Bank St.}]}] You'll not see any output in the browser, just the output in the console. But the code can be easily modified to do so. Once again, the complete Maven project can be downloaded here. Do you want to try accessing relational and non-relational (aka NoSQL) databases in the same PU ?

  Read the article

• What's up with LDoms: Part 2 - Creating a first, simple guest

  - by Stefan Hinker

  Welcome back! In the first part, we discussed the basic concepts of LDoms and how to configure a simple control domain.  We saw how resources were put aside for guest systems and what infrastructure we need for them.  With that, we are now ready to create a first, very simple guest domain.  In this first example, we'll keep things very simple.  Later on, we'll have a detailed look at things like sizing, IO redundancy, other types of IO as well as security. For now,let's start with this very simple guest.  It'll have one core's worth of CPU, one crypto unit, 8GB of RAM, a single boot disk and one network port.  CPU and RAM are easy.  The network port we'll create by attaching a virtual network port to the vswitch we created in the primary domain.  This is very much like plugging a cable into a computer system on one end and a network switch on the other.  For the boot disk, we'll need two things: A physical piece of storage to hold the data - this is called the backend device in LDoms speak.  And then a mapping between that storage and the guest domain, giving it access to that virtual disk.  For this example, we'll use a ZFS volume for the backend.  We'll discuss what other options there are for this and how to chose the right one in a later article.  Here we go: root@sun # ldm create mars root@sun # ldm set-vcpu 8 mars root@sun # ldm set-mau 1 mars root@sun # ldm set-memory 8g mars root@sun # zfs create rpool/guests root@sun # zfs create -V 32g rpool/guests/mars.bootdisk root@sun # ldm add-vdsdev /dev/zvol/dsk/rpool/guests/mars.bootdisk \ mars.root@primary-vds root@sun # ldm add-vdisk root mars.root@primary-vds mars root@sun # ldm add-vnet net0 switch-primary mars That's all, mars is now ready to power on.  There are just three commands between us and the OK prompt of mars:  We have to "bind" the domain, start it and connect to its console.  Binding is the process where the hypervisor actually puts all the pieces that we've configured together.  If we made a mistake, binding is where we'll be told (starting in version 2.1, a lot of sanity checking has been put into the config commands themselves, but binding will catch everything else).  Once bound, we can start (and of course later stop) the domain, which will trigger the boot process of OBP.  By default, the domain will then try to boot right away.  If we don't want that, we can set "auto-boot?" to false.  Finally, we'll use telnet to connect to the console of our newly created guest.  The output of "ldm list" shows us what port has been assigned to mars.  By default, the console service only listens on the loopback interface, so using telnet is not a large security concern here. root@sun # ldm set-variable auto-boot\?=false mars root@sun # ldm bind mars root@sun # ldm start mars root@sun # ldm list NAME STATE FLAGS CONS VCPU MEMORY UTIL UPTIME primary active -n-cv- UART 8 7680M 0.5% 1d 4h 30m mars active -t---- 5000 8 8G 12% 1s root@sun # telnet localhost 5000 Trying 127.0.0.1... Connected to localhost. Escape character is '^]'. ~Connecting to console "mars" in group "mars" .... Press ~? for control options .. {0} ok banner SPARC T3-4, No Keyboard Copyright (c) 1998, 2011, Oracle and/or its affiliates. All rights reserved. OpenBoot 4.33.1, 8192 MB memory available, Serial # 87203131. Ethernet address 0:21:28:24:1b:50, Host ID: 85241b50. {0} ok We're done, mars is ready to install Solaris, preferably using AI, of course ;-)  But before we do that, let's have a little look at the OBP environment to see how our virtual devices show up here: {0} ok printenv auto-boot? auto-boot? = false {0} ok printenv boot-device boot-device = disk net {0} ok devalias root /virtual-devices@100/channel-devices@200/disk@0 net0 /virtual-devices@100/channel-devices@200/network@0 net /virtual-devices@100/channel-devices@200/network@0 disk /virtual-devices@100/channel-devices@200/disk@0 virtual-console /virtual-devices/console@1 name aliases We can see that setting the OBP variable "auto-boot?" to false with the ldm command worked.  Of course, we'd normally set this to "true" to allow Solaris to boot right away once the LDom guest is started.  The setting for "boot-device" is the default "disk net", which means OBP would try to boot off the devices pointed to by the aliases "disk" and "net" in that order, which usually means "disk" once Solaris is installed on the disk image.  The actual devices these aliases point to are shown with the command "devalias".  Here, we have one line for both "disk" and "net".  The device paths speak for themselves.  Note that each of these devices has a second alias: "net0" for the network device and "root" for the disk device.  These are the very same names we've given these devices in the control domain with the commands "ldm add-vnet" and "ldm add-vdisk".  Remember this, as it is very useful once you have several dozen disk devices... To wrap this up, in this part we've created a simple guest domain, complete with CPU, memory, boot disk and network connectivity.  This should be enough to get you going.  I will cover all the more advanced features and a little more theoretical background in several follow-on articles.  For some background reading, I'd recommend the following links: LDoms 2.2 Admin Guide: Setting up Guest Domains Virtual Console Server: vntsd manpage - This includes the control sequences and commands available to control the console session. OpenBoot 4.x command reference - All the things you can do at the ok prompt

  Read the article

• Creating an SMF service for mercurial web server

  - by Chris W Beal

  I'm working on a project at the moment, which has a number of contributers. We're managing the project gate (which is stand alone) with mercurial. We want to have an easy way of seeing the changelog, so we can show management what is going on.  Luckily mercurial provides a basic web server which allows you to see the changes, and drill in to change sets. This can be run as a daemon, but as it was running on our build server, every time it was rebooted, someone needed to remember to start the process again. This is of course a classic usage of SMF. Now I'm not an experienced person at writing SMF services, so it took me 1/2 an hour or so to figure it out the first time. But going forward I should know what I'm doing a bit better. I did reference this doc extensively. Taking a step back, the command to start the mercurial web server is $ hg serve -p <port number> -d So we somehow need to get SMF to run that command for us. In the simplest form, SMF services are really made up of two components. The manifest Usually lives in /var/svc/manifest somewhere Can be imported from any location The method Usually live in /lib/svc/method I simply put the script straight in that directory. Not very repeatable, but it worked Can take an argument of start, stop, or refresh Lets start with the manifest. This looks pretty complex, but all it's doing is describing the service name, the dependencies, the start and stop methods, and some properties. The properties can be by instance, that is to say I could have multiple hg serve processes handling different mercurial projects, on different ports simultaneously Here is the manifest I wrote. I stole extensively from the examples in the Documentation. So my manifest looks like this $ cat hg-serve.xml <?xml version="1.0"?> <!DOCTYPE service_bundle SYSTEM "/usr/share/lib/xml/dtd/service_bundle.dtd.1"> <service_bundle type='manifest' name='hg-serve'> <service name='application/network/hg-serve' type='service' version='1'> <dependency name='network' grouping='require_all' restart_on='none' type='service'> <service_fmri value='svc:/milestone/network:default' /> </dependency> <exec_method type='method' name='start' exec='/lib/svc/method/hg-serve %m' timeout_seconds='2' /> <exec_method type='method' name='stop' exec=':kill' timeout_seconds='2'> </exec_method> <instance name='project-gate' enabled='true'> <method_context> <method_credential user='root' group='root' /> </method_context> <property_group name='hg-serve' type='application'> <propval name='path' type='astring' value='/src/project-gate'/> <propval name='port' type='astring' value='9998' /> </property_group> </instance> <stability value='Evolving' /> <template> <common_name> <loctext xml:lang='C'>hg-serve</loctext> </common_name> <documentation> <manpage title='hg' section='1' /> </documentation> </template> </service> </service_bundle> So the only things I had to decide on in this are the service name "application/network/hg-serve" the start and stop methods (more of which later) and the properties. This is the information I need to pass to the start method script. In my case the port I want to start the web server on "9998", and the path to the source gate "/src/project-gate". These can be read in to the start method. So now lets look at the method scripts $ cat /lib/svc/method/hg-serve #!/sbin/sh # # # Copyright (c) 2012, Oracle and/or its affiliates. All rights reserved. # # Standard prolog # . /lib/svc/share/smf_include.sh if [ -z $SMF_FMRI ]; then echo "SMF framework variables are not initialized." exit $SMF_EXIT_ERR fi # # Build the command line flags # # Get the port and directory from the SMF properties port=`svcprop -c -p hg-serve/port $SMF_FMRI` dir=`svcprop -c -p hg-serve/path $SMF_FMRI` echo "$1" case "$1" in 'start') cd $dir /usr/bin/hg serve -d -p $port ;; *) echo "Usage: $0 {start|refresh|stop}" exit 1 ;; esac exit $SMF_EXIT_OK This is all pretty self explanatory, we read the port and directory using svcprop, and use those simply to run a command in the start case. We don't need to implement a stop case, as the manifest says to use "exec=':kill'for the stop method. Now all we need to do is import the manifest and start the service, but first verify the manifest # svccfg verify /path/to/hg-serve.xml If that doesn't give an error try importing it # svccfg import /path/to/hg-serve.xml If like me you originally put the hg-serve.xml file in /var/svc/manifest somewhere you'll get an error and told to restart the import service svccfg: Restarting svc:/system/manifest-import The manifest being imported is from a standard location and should be imported with the command : svcadm restart svc:/system/manifest-import # svcadm restart svc:/system/manifest-import and you're nearly done. You can look at the service using svcs -l # svcs -l hg-serve fmri svc:/application/network/hg-serve:project-gate name hg-serve enabled false state disabled next_state none state_time Thu May 31 16:11:47 2012 logfile /var/svc/log/application-network-hg-serve:project-gate.log restarter svc:/system/svc/restarter:default contract_id 15749 manifest /var/svc/manifest/network/hg/hg-serve.xml dependency require_all/none svc:/milestone/network:default (online) And look at the interesting properties # svcprop hg-serve hg-serve/path astring /src/project-gate hg-serve/port astring 9998 ...stuff deleted.... Then simply enable the service and if every things gone right, you can point your browser at http://server:9998 and get a nice graphical log of project activity. # svcadm enable hg-serve # svcs -l hg-serve fmri svc:/application/network/hg-serve:project-gate name hg-serve enabled true state online next_state none state_time Thu May 31 16:18:11 2012 logfile /var/svc/log/application-network-hg-serve:project-gate.log restarter svc:/system/svc/restarter:default contract_id 15858 manifest /var/svc/manifest/network/hg/hg-serve.xml dependency require_all/none svc:/milestone/network:default (online) None of this is rocket science, but a bit fiddly. Hence I thought I'd blog it. It might just be you see this in google and it clicks with you more than one of the many other blogs or how tos about it. Plus I can always refer back to it myself in 3 weeks, when I want to add another project to the server, and I've forgotten how to do it.

  Read the article

• Das T5-4 TPC-H Ergebnis naeher betrachtet

  - by Stefan Hinker

  Inzwischen haben vermutlich viele das neue TPC-H Ergebnis der SPARC T5-4 gesehen, das am 7. Juni bei der TPC eingereicht wurde.  Die wesentlichen Punkte dieses Benchmarks wurden wie gewohnt bereits von unserer Benchmark-Truppe auf  "BestPerf" zusammengefasst.  Es gibt aber noch einiges mehr, das eine naehere Betrachtung lohnt. Skalierbarkeit Das TPC raet von einem Vergleich von TPC-H Ergebnissen in unterschiedlichen Groessenklassen ab.  Aber auch innerhalb der 3000GB-Klasse ist es interessant: SPARC T4-4 mit 4 CPUs (32 Cores mit 3.0 GHz) liefert 205,792 QphH. SPARC T5-4 mit 4 CPUs (64 Cores mit 3.6 GHz) liefert 409,721 QphH. Das heisst, es fehlen lediglich 1863 QphH oder 0.45% zu 100% Skalierbarkeit, wenn man davon ausgeht, dass die doppelte Anzahl Kerne das doppelte Ergebnis liefern sollte.  Etwas anspruchsvoller, koennte man natuerlich auch einen Faktor von 2.4 erwarten, wenn man die hoehere Taktrate mit beruecksichtigt.  Das wuerde die Latte auf 493901 QphH legen.  Dann waere die SPARC T5-4 bei 83%.  Damit stellt sich die Frage: Was hat hier nicht skaliert?  Vermutlich der Plattenspeicher!  Auch hier lohnt sich eine naehere Betrachtung: Plattenspeicher Im Bericht auf BestPerf und auch im Full Disclosure Report der TPC stehen einige interessante Details zum Plattenspeicher und der Konfiguration.   In der Konfiguration der SPARC T4-4 wurden 12 2540-M2 Arrays verwendet, die jeweils ca. 1.5 GB/s Durchsatz liefert, insgesamt also eta 18 GB/s.  Dabei waren die Arrays offensichtlich mit jeweils 2 Kabeln pro Array direkt an die 24 8GBit FC-Ports des Servers angeschlossen.  Mit den 2x 8GBit Ports pro Array koennte man so ein theoretisches Maximum von 2GB/s erreichen.  Tatsaechlich wurden 1.5GB/s geliefert, was so ziemlich dem realistischen Maximum entsprechen duerfte. Fuer den Lauf mit der SPARC T5-4 wurden doppelt so viele Platten verwendet.  Dafuer wurden die 2540-M2 Arrays mit je einem zusaetzlichen Plattentray erweitert.  Mit dieser Konfiguration wurde dann (laut BestPerf) ein Maximaldurchsatz von 33 GB/s erreicht - nicht ganz das doppelte des SPARC T4-4 Laufs.  Um tatsaechlich den doppelten Durchsatz (36 GB/s) zu liefern, haette jedes der 12 Arrays 3 GB/s ueber seine 4 8GBit Ports liefern muessen.  Im FDR stehen nur 12 dual-port FC HBAs, was die Verwendung der Brocade FC Switches erklaert: Es wurden alle 4 8GBit ports jedes Arrays an die Switches angeschlossen, die die Datenstroeme dann in die 24 16GBit HBA ports des Servers buendelten.  Das theoretische Maximum jedes Storage-Arrays waere nun 4 GB/s.  Wenn man jedoch den Protokoll- und "Realitaets"-Overhead mit einrechnet, sind die tatsaechlich gelieferten 2.75 GB/s gar nicht schlecht.  Mit diesen Zahlen im Hinterkopf ist die Verdopplung des SPARC T4-4 Ergebnisses eine gute Leistung - und gleichzeitig eine gute Erklaerung, warum nicht bis zum 2.4-fachen skaliert wurde. Nebenbei bemerkt: Weder die SPARC T4-4 noch die SPARC T5-4 hatten in der gemessenen Konfiguration irgendwelche Flash-Devices. Mitbewerb Seit die T4 Systeme auf dem Markt sind, bemuehen sich unsere Mitbewerber redlich darum, ueberall den Eindruck zu hinterlassen, die Leistung des SPARC CPU-Kerns waere weiterhin mangelhaft.  Auch scheinen sie ueberzeugt zu sein, dass (ueber)grosse Caches und hohe Taktraten die einzigen Schluessel zu echter Server Performance seien.  Wenn ich mir nun jedoch die oeffentlichen TPC-H Ergebnisse ansehe, sehe ich dies: TPC-H @3000GB, Non-Clustered Systems System QphH SPARC T5-4 3.6 GHz SPARC T5 4/64 – 2048 GB 409,721.8 SPARC T4-4 3.0 GHz SPARC T4 4/32 – 1024 GB 205,792.0 IBM Power 780 4.1 GHz POWER7 8/32 – 1024 GB 192,001.1 HP ProLiant DL980 G7 2.27 GHz Intel Xeon X7560 8/64 – 512 GB 162,601.7 Kurz zusammengefasst: Mit 32 Kernen (mit 3 GHz und 4MB L3 Cache), liefert die SPARC T4-4 mehr QphH@3000GB ab als IBM mit ihrer 32 Kern Power7 (bei 4.1 GHz und 32MB L3 Cache) und auch mehr als HP mit einem 64 Kern Intel Xeon System (2.27 GHz und 24MB L3 Cache).  Ich frage mich, wo genau SPARC hier mangelhaft ist? Nun koennte man natuerlich argumentieren, dass beide Ergebnisse nicht gerade neu sind.  Nun, in Ermangelung neuerer Ergebnisse kann man ja mal ein wenig spekulieren: IBMs aktueller Performance Report listet die o.g. IBM Power 780 mit einem rPerf Wert von 425.5.  Ein passendes Nachfolgesystem mit Power7+ CPUs waere die Power 780+ mit 64 Kernen, verfuegbar mit 3.72 GHz.  Sie wird mit einem rPerf Wert von  690.1 angegeben, also 1.62x mehr.  Wenn man also annimmt, dass Plattenspeicher nicht der limitierende Faktor ist (IBM hat mit 177 SSDs getestet, sie duerfen das gerne auf 400 erhoehen) und IBMs eigene Leistungsabschaetzung zugrunde legt, darf man ein theoretisches Ergebnis von 311398 QphH@3000GB erwarten.  Das waere dann allerdings immer noch weit von dem Ergebnis der SPARC T5-4 entfernt, und gerade in der von IBM so geschaetzen "per core" Metric noch weniger vorteilhaft. In der x86-Welt sieht es nicht besser aus.  Leider gibt es von Intel keine so praktischen rPerf-Tabellen.  Daher muss ich hier fuer eine Schaetzung auf SPECint_rate2006 zurueckgreifen.  (Ich bin kein grosser Fan von solchen Kreuz- und Querschaetzungen.  Insb. SPECcpu ist nicht besonders geeignet, um Datenbank-Leistung abzuschaetzen, da fast kein IO im Spiel ist.)  Das o.g. HP System wird bei SPEC mit 1580 CINT2006_rate gelistet.  Das bis einschl. 2013-06-14 beste Resultat fuer den neuen Intel Xeon E7-4870 mit 8 CPUs ist 2180 CINT2006_rate.  Das ist immerhin 1.38x besser.  (Wenn man nur die Taktrate beruecksichtigen wuerde, waere man bei 1.32x.)  Hier weiter zu rechnen, ist muessig, aber fuer die ungeduldigen Leser hier eine kleine tabellarische Zusammenfassung: TPC-H @3000GB Performance Spekulationen System QphH* Verbesserung gegenueber der frueheren Generation SPARC T4-4 32 cores SPARC T4 205,792 2x SPARC T5-464 cores SPARC T5 409,721 IBM Power 780 32 cores Power7 192,001 1.62x IBM Power 780+ 64 cores Power7+  311,398* HP ProLiant DL980 G764 cores Intel Xeon X7560 162,601 1.38x HP ProLiant DL980 G780 cores Intel Xeon E7-4870    224,348* * Keine echten Resultate  - spekulative Werte auf der Grundlage von rPerf (Power7+) oder SPECint_rate2006 (HP) Natuerlich sind IBM oder HP herzlich eingeladen, diese Werte zu widerlegen.  Aber stand heute warte ich noch auf aktuelle Benchmark Veroffentlichungen in diesem Datensegment. Was koennen wir also zusammenfassen? Es gibt einige Hinweise, dass der Plattenspeicher der begrenzende Faktor war, der die SPARC T5-4 daran hinderte, auf jenseits von 2x zu skalieren Der Mythos, dass SPARC Kerne keine Leistung bringen, ist genau das - ein Mythos.  Wie sieht es umgekehrt eigentlich mit einem TPC-H Ergebnis fuer die Power7+ aus? Cache ist nicht der magische Performance-Schalter, fuer den ihn manche Leute offenbar halten. Ein System, eine CPU-Architektur und ein Betriebsystem jenseits einer gewissen Grenze zu skalieren ist schwer.  In der x86-Welt scheint es noch ein wenig schwerer zu sein. Was fehlt?  Nun, das Thema Preis/Leistung ueberlasse ich gerne den Verkaeufern ;-) Und zu guter Letzt: Nein, ich habe mich nicht ins Marketing versetzen lassen.  Aber manchmal kann ich mich einfach nicht zurueckhalten... Disclosure Statements The views expressed on this blog are my own and do not necessarily reflect the views of Oracle. TPC-H, QphH, $/QphH are trademarks of Transaction Processing Performance Council (TPC). For more information, see www.tpc.org, results as of 6/7/13. Prices are in USD. SPARC T5-4 409,721.8 QphH@3000GB, $3.94/QphH@3000GB, available 9/24/13, 4 processors, 64 cores, 512 threads; SPARC T4-4 205,792.0 QphH@3000GB, $4.10/QphH@3000GB, available 5/31/12, 4 processors, 32 cores, 256 threads; IBM Power 780 QphH@3000GB, 192,001.1 QphH@3000GB, $6.37/QphH@3000GB, available 11/30/11, 8 processors, 32 cores, 128 threads; HP ProLiant DL980 G7 162,601.7 QphH@3000GB, $2.68/QphH@3000GB available 10/13/10, 8 processors, 64 cores, 128 threads. SPEC and the benchmark names SPECfp and SPECint are registered trademarks of the Standard Performance Evaluation Corporation. Results as of June 18, 2013 from www.spec.org. HP ProLiant DL980 G7 (2.27 GHz, Intel Xeon X7560): 1580 SPECint_rate2006; HP ProLiant DL980 G7 (2.4 GHz, Intel Xeon E7-4870): 2180 SPECint_rate2006,

  Read the article

• T4 Performance Counters explained

  - by user13346607

  Now that T4 is out for a few month some people might have wondered what details of the new pipeline you can monitor. A "cpustat -h" lists a lot of events that can be monitored, and only very few are self-explanatory. I will try to give some insight on all of them, some of these "PIC events" require an in-depth knowledge of T4 pipeline. Over time I will try to explain these, for the time being these events should simply be ignored. (Side note: some counters changed from tape-out 1.1 (*only* used in the T4 beta program) to tape-out 1.2 (used in the systems shipping today) The table only lists the tape-out 1.2 counters) 0 0 1 1058 6033 Oracle Microelectronics 50 14 7077 14.0 Normal 0 false false false EN-US JA X-NONE /* Style Definitions */ table.MsoNormalTable {mso-style-name:"Table Normal"; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-style-noshow:yes; mso-style-priority:99; mso-style-parent:""; mso-padding-alt:0cm 5.4pt 0cm 5.4pt; mso-para-margin:0cm; mso-para-margin-bottom:.0001pt; mso-pagination:widow-orphan; font-size:12.0pt; font-family:Cambria; mso-ascii-font-family:Cambria; mso-ascii-theme-font:minor-latin; mso-hansi-font-family:Cambria; mso-hansi-theme-font:minor-latin;} pic name (cpustat) Prose Comment Sel-pipe-drain-cycles, Sel-0-[wait|ready], Sel-[1,2] Sel-0-wait counts cycles a strand waits to be selected. Some reasons can be counted in detail; these are: Sel-0-ready: Cycles a strand was ready but not selected, that can signal pipeline oversubscription Sel-1: Cycles only one instruction or µop was selected Sel-2: Cycles two instructions or µops were selected Sel-pipe-drain-cycles: cf. PRM footnote 8 to table 10.2 Pick-any, Pick-[0|1|2|3] Cycles one, two, three, no or at least one instruction or µop is picked Instr_FGU_crypto Number of FGU or crypto instructions executed on that vcpu Instr_ld dto. for load Instr_st dto. for store SPR_ring_ops dto. for SPR ring ops Instr_other dto. for all other instructions not listed above, PRM footnote 7 to table 10.2 lists the instructions Instr_all total number of instructions executed on that vcpu Sw_count_intr Nr of S/W count instructions on that vcpu (sethi %hi(fc000),%g0 (whatever that is))  Atomics nr of atomic ops, which are LDSTUB/a, CASA/XA, and SWAP/A SW_prefetch Nr of PREFETCH or PREFETCHA instructions Block_ld_st Block loads or store on that vcpu IC_miss_nospec, IC_miss_[L2_or_L3|local|remote]\ _hit_nospec Various I$ misses, distinguished by where they hit. All of these count per thread, but only primary events: T4 counts only the first occurence of an I$ miss on a core for a certain instruction. If one strand misses in I$ this miss is counted, but if a second strand on the same core misses while the first miss is being resolved, that second miss is not counted This flavour of I$ misses counts only misses that are caused by instruction that really commit (note the "_nospec") BTC_miss Branch target cache miss ITLB_miss ITLB misses (synchronously counted) ITLB_miss_asynch dto. but asynchronously [I|D]TLB_fill_\ [8KB|64KB|4MB|256MB|2GB|trap] H/W tablewalk events that fill ITLB or DTLB with translation for the corresponding page size. The “_trap” event occurs if the HWTW was not able to fill the corresponding TLB IC_mtag_miss, IC_mtag_miss_\ [ptag_hit|ptag_miss|\ ptag_hit_way_mismatch] I$ micro tag misses, with some options for drill down Fetch-0, Fetch-0-all fetch-0 counts nr of cycles nothing was fetched for this particular strand, fetch-0-all counts cycles nothing was fetched for all strands on a core Instr_buffer_full Cycles the instruction buffer for a strand was full, thereby preventing any fetch BTC_targ_incorrect Counts all occurences of wrongly predicted branch targets from the BTC [PQ|ROB|LB|ROB_LB|SB|\ ROB_SB|LB_SB|RB_LB_SB|\ DTLB_miss]\ _tag_wait ST_q_tag_wait is listed under sl=20. These counters monitor pipeline behaviour therefore they are not strand specific: PQ_...: cycles Rename stage waits for a Pick Queue tag (might signal memory bound workload for single thread mode, cf. Mail from Richard Smith) ROB_...: cycles Select stage waits for a ROB (ReOrderBuffer) tag LB_...: cycles Select stage waits for a Load Buffer tag SB_...: cycles Select stage waits for Store Buffer tag combinations of the above are allowed, although some of these events can overlap, the counter will only be incremented once per cycle if any of these occur DTLB_...: cycles load or store instructions wait at Pick stage for a DTLB miss tag [ID]TLB_HWTW_\ [L2_hit|L3_hit|L3_miss|all] Counters for HWTW accesses caused by either DTLB or ITLB misses. Canbe further detailed by where they hit IC_miss_L2_L3_hit, IC_miss_local_remote_remL3_hit, IC_miss I$ prefetches that were dropped because they either miss in L2$ or L3$ This variant counts misses regardless if the causing instruction commits or not DC_miss_nospec, DC_miss_[L2_L3|local|remote_L3]\ _hit_nospec D$ misses either in general or detailed by where they hit cf. the explanation for the IC_miss in two flavours for an explanation of _nospec and the reasoning for two DC_miss counters DTLB_miss_asynch counts all DTLB misses asynchronously, there is no way to count them synchronously DC_pref_drop_DC_hit, SW_pref_drop_[DC_hit|buffer_full] L1-D$ h/w prefetches that were dropped because of a D$ hit, counted per core. The others count software prefetches per strand [Full|Partial]_RAW_hit_st_[buf|q] Count events where a load wants to get data that has not yet been stored, i. e. it is still inside the pipeline. The data might be either still in the store buffer or in the store queue. If the load's data matches in the SB and in the store queue the data in buffer takes precedence of course since it is younger [IC|DC]_evict_invalid, [IC|DC|L1]_snoop_invalid, [IC|DC|L1]_invalid_all Counter for invalidated cache evictions per core St_q_tag_wait Number of cycles pipeline waits for a store queue tag, of course counted per core Data_pref_[drop_L2|drop_L3|\ hit_L2|hit_L3|\ hit_local|hit_remote] Data prefetches that can be further detailed by either why they were dropped or where they did hit St_hit_[L2|L3], St_L2_[local|remote]_C2C, St_local, St_remote Store events distinguished by where they hit or where they cause a L2 cache-to-cache transfer, i.e. either a transfer from another L2$ on the same die or from a different die DC_miss, DC_miss_\ [L2_L3|local|remote]_hit D$ misses either in general or detailed by where they hit cf. the explanation for the IC_miss in two flavours for an explanation of _nospec and the reasoning for two DC_miss counters L2_[clean|dirty]_evict Per core clean or dirty L2$ evictions L2_fill_buf_full, L2_wb_buf_full, L2_miss_buf_full Per core L2$ buffer events, all count number of cycles that this state was present L2_pipe_stall Per core cycles pipeline stalled because of L2$ Branches Count branches (Tcc, DONE, RETRY, and SIT are not counted as branches) Br_taken Counts taken branches (Tcc, DONE, RETRY, and SIT are not counted as branches) Br_mispred, Br_dir_mispred, Br_trg_mispred, Br_trg_mispred_\ [far_tbl|indir_tbl|ret_stk] Counter for various branch misprediction events.  Cycles_user counts cycles, attribute setting hpriv, nouser, sys controls addess space to count in Commit-[0|1|2], Commit-0-all, Commit-1-or-2 Number of times either no, one, or two µops commit for a strand. Commit-0-all counts number of times no µop commits for the whole core, cf. footnote 11 to table 10.2 in PRM for a more detailed explanation on how this counters interacts with the privilege levels

  Read the article

• Wildcards!

  - by Tim Dexter

  Yes, its been a while, Im sorry, mumble, mumble ... no excuses. Well other than its been, as my son would say 'hecka busy.' On a brighter note I see Kan has been posting some cool stuff in my absence, long may he continue! I received a question today asking about using a wildcard in a template, something like: <?if:INVOICE = 'MLP*'?> where * is the wildcard Well that particular try does not work but you can do it without building your own wildcard function. XSL, the underpinning language of the RTF templates, has some useful string functions - you can find them listed here. I used the starts-with function to achieve a simple wildcard scenario but the contains can be used in conjunction with some of the others to build something more sophisticated. Assume I have a a list of friends and the amounts of money they owe me ... Im very generous and my interest rates a pretty competitive :0) <ROWSET> <ROW> <NAME>Andy</NAME> <AMT>100</AMT> </ROW> <ROW> <NAME>Andrew</NAME> <AMT>60</AMT> </ROW> <ROW> <NAME>Aaron</NAME> <AMT>50</AMT> </ROW> <ROW> <NAME>Alice</NAME> <AMT>40</AMT> </ROW> <ROW> <NAME>Bob</NAME> <AMT>10</AMT> </ROW> <ROW> <NAME>Bill</NAME> <AMT>100</AMT> </ROW> Now, listing my friends is easy enough <for-each:ROW> <NAME> <AMT> <end for-each> but lets say I just want to see all my friends beginning with 'A'. To do that I can use an XPATH expression to filter the data and tack it on to the for-each expression. This is more efficient that using an 'if' statement just inside the for-each. <?for-each:ROW[starts-with(NAME,'A')]?> will find me all the A's. The square braces denote the start of the XPATH expression. starts-with is the function Im calling and Im passing the value I want to check i.e. NAME and the string Im looking for. Just substitute in the characters you are looking for. You can of course use the function in a if statement too. <?if:starts-with(NAME,'A')?><?attribute@incontext:color;'red'?><?end if?> Notice I removed the square braces, this will highlight text red if the name begins with an 'A' You can even use the function to do conditional calculations: <?sum (AMT[starts-with(../NAME,'A')])?> Sum only the amounts where the name begins with an 'A' Notice the square braces are back, its a function we want to apply to the AMT field. Also notice that we need to use ../NAME. The AMT and NAME elements are at the same level in the tree, so when we are at the AMT level we need the ../ to go up a level to then come back down to test the NAME value. I have built out the above functions in a sample template here. Huge prizes for the first person to come up with a 'true' wildcard solution i.e. if NAME like '*im*exter* demand cash now!

  Read the article

• The Faces in the Crowdsourcing

  - by Applications User Experience

  By Jeff Sauro, Principal Usability Engineer, Oracle Imagine having access to a global workforce of hundreds of thousands of people who can perform tasks or provide feedback on a design quickly and almost immediately. Distributing simple tasks not easily done by computers to the masses is called "crowdsourcing" and until recently was an interesting concept, but due to practical constraints wasn't used often. Enter Amazon.com. For five years, Amazon has hosted a service called Mechanical Turk, which provides an easy interface to the crowds. The service has almost half a million registered, global users performing a quarter of a million human intelligence tasks (HITs). HITs are submitted by individuals and companies in the U.S. and pay from $.01 for simple tasks (such as determining if a picture is offensive) to several dollars (for tasks like transcribing audio). What do we know about the people who toil away in this digital crowd? Can we rely on the work done in this anonymous marketplace? A rendering of the actual Mechanical Turk (from Wikipedia) Knowing who is behind Amazon's Mechanical Turk is fitting, considering the history of the actual Mechanical Turk. In the late 1800's, a mechanical chess-playing machine awed crowds as it beat master chess players in what was thought to be a mechanical miracle. It turned out that the creator, Wolfgang von Kempelen, had a small person (also a chess master) hiding inside the machine operating the arms to provide the illusion of automation. The field of human computer interaction (HCI) is quite familiar with gathering user input and incorporating it into all stages of the design process. It makes sense then that Mechanical Turk was a popular discussion topic at the recent Computer Human Interaction usability conference sponsored by the Association for Computing Machinery in Atlanta. It is already being used as a source for input on Web sites (for example, Feedbackarmy.com) and behavioral research studies. Two papers shed some light on the faces in this crowd. One paper tells us about the shifting demographics from mostly stay-at-home moms to young men in India. The second paper discusses the reliability and quality of work from the workers. Just who exactly would spend time doing tasks for pennies? In "Who are the crowdworkers?" University of California researchers Ross, Silberman, Zaldivar and Tomlinson conducted a survey of Mechanical Turk worker demographics and compared it to a similar survey done two years before. The initial survey reported workers consisting largely of young, well-educated women living in the U.S. with annual household incomes above $40,000. The more recent survey reveals a shift in demographics largely driven by an influx of workers from India. Indian workers went from 5% to over 30% of the crowd, and this block is largely male (two-thirds) with a higher average education than U.S. workers, and 64% report an annual income of less than $10,000 (keeping in mind $1 has a lot more purchasing power in India). This shifting demographic certainly has implications as language and culture can play critical roles in the outcome of HITs. Of course, the demographic data came from paying Turkers $.10 to fill out a survey, so there is some question about both a self-selection bias (characteristics which cause Turks to take this survey may be unrepresentative of the larger population), not to mention whether we can really trust the data we get from the crowd. Crowds can perform tasks or provide feedback on a design quickly and almost immediately for usability testing. (Photo attributed to victoriapeckham Flikr While having immediate access to a global workforce is nice, one major problem with Mechanical Turk is the incentive structure. Individuals and companies that deploy HITs want quality responses for a low price. Workers, on the other hand, want to complete the task and get paid as quickly as possible, so that they can get on to the next task. Since many HITs on Mechanical Turk are surveys, how valid and reliable are these results? How do we know whether workers are just rushing through the multiple-choice responses haphazardly answering? In "Are your participants gaming the system?" researchers at Carnegie Mellon (Downs, Holbrook, Sheng and Cranor) set up an experiment to find out what percentage of their workers were just in it for the money. The authors set up a 30-minute HIT (one of the more lengthy ones for Mechanical Turk) and offered a very high $4 to those who qualified and $.20 to those who did not. As part of the HIT, workers were asked to read an email and respond to two questions that determined whether workers were likely rushing through the HIT and not answering conscientiously. One question was simple and took little effort, while the second question required a bit more work to find the answer. Workers were led to believe other factors than these two questions were the qualifying aspect of the HIT. Of the 2000 participants, roughly 1200 (or 61%) answered both questions correctly. Eighty-eight percent answered the easy question correctly, and 64% answered the difficult question correctly. In other words, about 12% of the crowd were gaming the system, not paying enough attention to the question or making careless errors. Up to about 40% won't put in more than a modest effort to get paid for a HIT. Young men and those that considered themselves in the financial industry tended to be the most likely to try to game the system. There wasn't a breakdown by country, but given the demographic information from the first article, we could infer that many of these young men come from India, which makes language and other cultural differences a factor. These articles raise questions about the role of crowdsourcing as a means for getting quick user input at low cost. While compensating users for their time is nothing new, the incentive structure and anonymity of Mechanical Turk raises some interesting questions. How complex of a task can we ask of the crowd, and how much should these workers be paid? Can we rely on the information we get from these professional users, and if so, how can we best incorporate it into designing more usable products? Traditional usability testing will still play a central role in enterprise software. Crowdsourcing doesn't replace testing; instead, it makes certain parts of gathering user feedback easier. One can turn to the crowd for simple tasks that don't require specialized skills and get a lot of data fast. As more studies are conducted on Mechanical Turk, I suspect we will see crowdsourcing playing an increasing role in human computer interaction and enterprise computing. References: Downs, J. S., Holbrook, M. B., Sheng, S., and Cranor, L. F. 2010. Are your participants gaming the system?: screening mechanical turk workers. In Proceedings of the 28th international Conference on Human Factors in Computing Systems (Atlanta, Georgia, USA, April 10 - 15, 2010). CHI '10. ACM, New York, NY, 2399-2402. Link: http://doi.acm.org/10.1145/1753326.1753688 Ross, J., Irani, L., Silberman, M. S., Zaldivar, A., and Tomlinson, B. 2010. Who are the crowdworkers?: shifting demographics in mechanical turk. In Proceedings of the 28th of the international Conference Extended Abstracts on Human Factors in Computing Systems (Atlanta, Georgia, USA, April 10 - 15, 2010). CHI EA '10. ACM, New York, NY, 2863-2872. Link: http://doi.acm.org/10.1145/1753846.1753873

  Read the article

• NUMA-aware placement of communication variables

  - by Dave

  For classic NUMA-aware programming I'm typically most concerned about simple cold, capacity and compulsory misses and whether we can satisfy the miss by locally connected memory or whether we have to pull the line from its home node over the coherent interconnect -- we'd like to minimize channel contention and conserve interconnect bandwidth. That is, for this style of programming we're quite aware of where memory is homed relative to the threads that will be accessing it. Ideally, a page is collocated on the node with the thread that's expected to most frequently access the page, as simple misses on the page can be satisfied without resorting to transferring the line over the interconnect. The default "first touch" NUMA page placement policy tends to work reasonable well in this regard. When a virtual page is first accessed, the operating system will attempt to provision and map that virtual page to a physical page allocated from the node where the accessing thread is running. It's worth noting that the node-level memory interleaving granularity is usually a multiple of the page size, so we can say that a given page P resides on some node N. That is, the memory underlying a page resides on just one node. But when thinking about accesses to heavily-written communication variables we normally consider what caches the lines underlying such variables might be resident in, and in what states. We want to minimize coherence misses and cache probe activity and interconnect traffic in general. I don't usually give much thought to the location of the home NUMA node underlying such highly shared variables. On a SPARC T5440, for instance, which consists of 4 T2+ processors connected by a central coherence hub, the home node and placement of heavily accessed communication variables has very little impact on performance. The variables are frequently accessed so likely in M-state in some cache, and the location of the home node is of little consequence because a requester can use cache-to-cache transfers to get the line. Or at least that's what I thought. Recently, though, I was exploring a simple shared memory point-to-point communication model where a client writes a request into a request mailbox and then busy-waits on a response variable. It's a simple example of delegation based on message passing. The server polls the request mailbox, and having fetched a new request value, performs some operation and then writes a reply value into the response variable. As noted above, on a T5440 performance is insensitive to the placement of the communication variables -- the request and response mailbox words. But on a Sun/Oracle X4800 I noticed that was not the case and that NUMA placement of the communication variables was actually quite important. For background an X4800 system consists of 8 Intel X7560 Xeons . Each package (socket) has 8 cores with 2 contexts per core, so the system is 8x8x2. Each package is also a NUMA node and has locally attached memory. Every package has 3 point-to-point QPI links for cache coherence, and the system is configured with a twisted ladder "mobius" topology. The cache coherence fabric is glueless -- there's not central arbiter or coherence hub. The maximum distance between any two nodes is just 2 hops over the QPI links. For any given node, 3 other nodes are 1 hop distant and the remaining 4 nodes are 2 hops distant. Using a single request (client) thread and a single response (server) thread, a benchmark harness explored all permutations of NUMA placement for the two threads and the two communication variables, measuring the average round-trip-time and throughput rate between the client and server. In this benchmark the server simply acts as a simple transponder, writing the request value plus 1 back into the reply field, so there's no particular computation phase and we're only measuring communication overheads. In addition to varying the placement of communication variables over pairs of nodes, we also explored variations where both variables were placed on one page (and thus on one node) -- either on the same cache line or different cache lines -- while varying the node where the variables reside along with the placement of the threads. The key observation was that if the client and server threads were on different nodes, then the best placement of variables was to have the request variable (written by the client and read by the server) reside on the same node as the client thread, and to place the response variable (written by the server and read by the client) on the same node as the server. That is, if you have a variable that's to be written by one thread and read by another, it should be homed with the writer thread. For our simple client-server model that means using split request and response communication variables with unidirectional message flow on a given page. This can yield up to twice the throughput of less favorable placement strategies. Our X4800 uses the QPI 1.0 protocol with source-based snooping. Briefly, when node A needs to probe a cache line it fires off snoop requests to all the nodes in the system. Those recipients then forward their response not to the original requester, but to the home node H of the cache line. H waits for and collects the responses, adjudicates and resolves conflicts and ensures memory-model ordering, and then sends a definitive reply back to the original requester A. If some node B needed to transfer the line to A, it will do so by cache-to-cache transfer and let H know about the disposition of the cache line. A needs to wait for the authoritative response from H. So if a thread on node A wants to write a value to be read by a thread on node B, the latency is dependent on the distances between A, B, and H. We observe the best performance when the written-to variable is co-homed with the writer A. That is, we want H and A to be the same node, as the writer doesn't need the home to respond over the QPI link, as the writer and the home reside on the very same node. With architecturally informed placement of communication variables we eliminate at least one QPI hop from the critical path. Newer Intel processors use the QPI 1.1 coherence protocol with home-based snooping. As noted above, under source-snooping a requester broadcasts snoop requests to all nodes. Those nodes send their response to the home node of the location, which provides memory ordering, reconciles conflicts, etc., and then posts a definitive reply to the requester. In home-based snooping the snoop probe goes directly to the home node and are not broadcast. The home node can consult snoop filters -- if present -- and send out requests to retrieve the line if necessary. The 3rd party owner of the line, if any, can respond either to the home or the original requester (or even to both) according to the protocol policies. There are myriad variations that have been implemented, and unfortunately vendor terminology doesn't always agree between vendors or with the academic taxonomy papers. The key is that home-snooping enables the use of a snoop filter to reduce interconnect traffic. And while home-snooping might have a longer critical path (latency) than source-based snooping, it also may require fewer messages and less overall bandwidth. It'll be interesting to reprise these experiments on a platform with home-based snooping. While collecting data I also noticed that there are placement concerns even in the seemingly trivial case when both threads and both variables reside on a single node. Internally, the cores on each X7560 package are connected by an internal ring. (Actually there are multiple contra-rotating rings). And the last-level on-chip cache (LLC) is partitioned in banks or slices, which with each slice being associated with a core on the ring topology. A hardware hash function associates each physical address with a specific home bank. Thus we face distance and topology concerns even for intra-package communications, although the latencies are not nearly the magnitude we see inter-package. I've not seen such communication distance artifacts on the T2+, where the cache banks are connected to the cores via a high-speed crossbar instead of a ring -- communication latencies seem more regular.

  Read the article

• BPM 11g - Dynamic Task Assignment with Multi-level Organization Units

  - by Mark Foster

  I've seen several requirements to have a more granular level of task assignment in BPM 11g based on some value in the data passed to the process. Parametric Roles is normally the first port of call to try to satisfy this requirement, but in this blog we will show how a lot of use-cases can be satisfied by the easier to implement and flexible Organization Unit. The Use-Case Task assignment is to an approval group containing several users. At runtime, a location value in the input data determines which of the particular users the task is ultimately assigned to. In this case we use the Demo Community referenced in the SOA Admin Guide, and specifically the "LoanAnalyticGroup" which contains three users; "szweig", "mmitch" & "fkafka". In our scenario we would like to assign a task to "szweig" if the input data specifies that the location is "JapanCentral", to "fkafka" if the location is "JapanNorth" and to "mmitch" if "JapanSouth", and to all of them if the location is "Japan" i.e....   The Process Simple one human task process.... In the output data association of the "Start" activity we need to set the value of the "Organization Unit" predefined variable based on the input data (note that the  predefined variables can only be set on output data associations)....  ...and in the output data association of the human activity we will reset the "Organization Unit" to empty, always good practice to ensure that the Organization Unit will not be used for any subsequent human activities for which we do not require it.... Set Up the Organization Unit  Log in to the BPM Workspace with an administrator user (weblogic/welcome1 in our case) and choose the "Administration" option. Within "Roles" assign the "ProcessOwner" swim-lane for our process to "LoanAnalyticGroup".... Within "Organization Units" we can model our organization.... "Root Organization Unit" as "Japan" and "Child Organization Unit" as "Central", "South" & "North" as shown. As described previously, add user "szweig" to "Central", "mmitch" to "South" and "fkafka" to "North"....   Test the Process Invalid Data  First let us test with invalid data in the input to see what the consequences are, here we use "X" as input.... ...and looking at the instance we can see it has errored.... Organization Unit Root Level Assignment  Now let us see what happens if we have "Japan" in the input data.... ...looking in the "flow trace" we can see that the task has been assigned....  ... but who has the task been assigned to ? Let us look in the BPM Workspace for user "szweig"....  ...and for "mmitch"....  ... and for "fkafka"....  ...so we can see that with an Organization Unit at "Root" level we have successfully assigned the task to all users. Organization Unit Child Level Assignment  Now let us test with "Japan/North" in the input data.... ...and looking in "fkafka" workspace we see the task has been assigned, remember, he was associated with "JapanNorth"....   ... but what about the workspace of "szweig"....  ...no tasks assigned, neither has "mmitch", just as we expected. Summary  We have seen in this blog how to easily implement multi-level dynamic task routing using Organization Units, a common use-case and a simpler solution than Parametric Roles. 

  Read the article

• DTracing TCP congestion control

  - by user12820842

  In a previous post, I showed how we can use DTrace to probe TCP receive and send window events. TCP receive and send windows are in effect both about flow-controlling how much data can be received - the receive window reflects how much data the local TCP is prepared to receive, while the send window simply reflects the size of the receive window of the peer TCP. Both then represent flow control as imposed by the receiver. However, consider that without the sender imposing flow control, and a slow link to a peer, TCP will simply fill up it's window with sent segments. Dealing with multiple TCP implementations filling their peer TCP's receive windows in this manner, busy intermediate routers may drop some of these segments, leading to timeout and retransmission, which may again lead to drops. This is termed congestion, and TCP has multiple congestion control strategies. We can see that in this example, we need to have some way of adjusting how much data we send depending on how quickly we receive acknowledgement - if we get ACKs quickly, we can safely send more segments, but if acknowledgements come slowly, we should proceed with more caution. More generally, we need to implement flow control on the send side also. Slow Start and Congestion Avoidance From RFC2581, let's examine the relevant variables: "The congestion window (cwnd) is a sender-side limit on the amount of data the sender can transmit into the network before receiving an acknowledgment (ACK). Another state variable, the slow start threshold (ssthresh), is used to determine whether the slow start or congestion avoidance algorithm is used to control data transmission" Slow start is used to probe the network's ability to handle transmission bursts both when a connection is first created and when retransmission timers fire. The latter case is important, as the fact that we have effectively lost TCP data acts as a motivator for re-probing how much data the network can handle from the sending TCP. The congestion window (cwnd) is initialized to a relatively small value, generally a low multiple of the sending maximum segment size. When slow start kicks in, we will only send that number of bytes before waiting for acknowledgement. When acknowledgements are received, the congestion window is increased in size until cwnd reaches the slow start threshold ssthresh value. For most congestion control algorithms the window increases exponentially under slow start, assuming we receive acknowledgements. We send 1 segment, receive an ACK, increase the cwnd by 1 MSS to 2*MSS, send 2 segments, receive 2 ACKs, increase the cwnd by 2*MSS to 4*MSS, send 4 segments etc. When the congestion window exceeds the slow start threshold, congestion avoidance is used instead of slow start. During congestion avoidance, the congestion window is generally updated by one MSS for each round-trip-time as opposed to each ACK, and so cwnd growth is linear instead of exponential (we may receive multiple ACKs within a single RTT). This continues until congestion is detected. If a retransmit timer fires, congestion is assumed and the ssthresh value is reset. It is reset to a fraction of the number of bytes outstanding (unacknowledged) in the network. At the same time the congestion window is reset to a single max segment size. Thus, we initiate slow start until we start receiving acknowledgements again, at which point we can eventually flip over to congestion avoidance when cwnd ssthresh. Congestion control algorithms differ most in how they handle the other indication of congestion - duplicate ACKs. A duplicate ACK is a strong indication that data has been lost, since they often come from a receiver explicitly asking for a retransmission. In some cases, a duplicate ACK may be generated at the receiver as a result of packets arriving out-of-order, so it is sensible to wait for multiple duplicate ACKs before assuming packet loss rather than out-of-order delivery. This is termed fast retransmit (i.e. retransmit without waiting for the retransmission timer to expire). Note that on Oracle Solaris 11, the congestion control method used can be customized. See here for more details. In general, 3 or more duplicate ACKs indicate packet loss and should trigger fast retransmit . It's best not to revert to slow start in this case, as the fact that the receiver knew it was missing data suggests it has received data with a higher sequence number, so we know traffic is still flowing. Falling back to slow start would be excessive therefore, so fast recovery is used instead. Observing slow start and congestion avoidance The following script counts TCP segments sent when under slow start (cwnd ssthresh). #!/usr/sbin/dtrace -s #pragma D option quiet tcp:::connect-request / start[args[1]-cs_cid] == 0/ { start[args[1]-cs_cid] = 1; } tcp:::send / start[args[1]-cs_cid] == 1 && args[3]-tcps_cwnd tcps_cwnd_ssthresh / { @c["Slow start", args[2]-ip_daddr, args[4]-tcp_dport] = count(); } tcp:::send / start[args[1]-cs_cid] == 1 && args[3]-tcps_cwnd args[3]-tcps_cwnd_ssthresh / { @c["Congestion avoidance", args[2]-ip_daddr, args[4]-tcp_dport] = count(); } As we can see the script only works on connections initiated since it is started (using the start[] associative array with the connection ID as index to set whether it's a new connection (start[cid] = 1). From there we simply differentiate send events where cwnd ssthresh (congestion avoidance). Here's the output taken when I accessed a YouTube video (where rport is 80) and from an FTP session where I put a large file onto a remote system. # dtrace -s tcp_slow_start.d ^C ALGORITHM RADDR RPORT #SEG Slow start 10.153.125.222 20 6 Slow start 138.3.237.7 80 14 Slow start 10.153.125.222 21 18 Congestion avoidance 10.153.125.222 20 1164 We see that in the case of the YouTube video, slow start was exclusively used. Most of the segments we sent in that case were likely ACKs. Compare this case - where 14 segments were sent using slow start - to the FTP case, where only 6 segments were sent before we switched to congestion avoidance for 1164 segments. In the case of the FTP session, the FTP data on port 20 was predominantly sent with congestion avoidance in operation, while the FTP session relied exclusively on slow start. For the default congestion control algorithm - "newreno" - on Solaris 11, slow start will increase the cwnd by 1 MSS for every acknowledgement received, and by 1 MSS for each RTT in congestion avoidance mode. Different pluggable congestion control algorithms operate slightly differently. For example "highspeed" will update the slow start cwnd by the number of bytes ACKed rather than the MSS. And to finish, here's a neat oneliner to visually display the distribution of congestion window values for all TCP connections to a given remote port using a quantization. In this example, only port 80 is in use and we see the majority of cwnd values for that port are in the 4096-8191 range. # dtrace -n 'tcp:::send { @q[args[4]-tcp_dport] = quantize(args[3]-tcps_cwnd); }' dtrace: description 'tcp:::send ' matched 10 probes ^C 80 value ------------- Distribution ------------- count -1 | 0 0 |@@@@@@ 5 1 | 0 2 | 0 4 | 0 8 | 0 16 | 0 32 | 0 64 | 0 128 | 0 256 | 0 512 | 0 1024 | 0 2048 |@@@@@@@@@ 8 4096 |@@@@@@@@@@@@@@@@@@@@@@@@@@ 23 8192 | 0

  Read the article

• "Yes, but that's niche."

  - by Geertjan

  JavaOne 2012 has come to an end though it feels like it hasn't even started yet! What happened, time is a weird thing. Too many things to report on. James Gosling's appearance at the JavaOne community keynote was seen, by everyone (which is quite a lot) of people I talked to, as the highlight of the conference. It was interesting that the software for the Duke's Choice Award winning Liquid Robotics that James Gosling is now part of and came to talk about is a Swing application that uses the WorldWind libraries. It was also interesting that James Gosling pointed out to the conference: "There are things you can't do using HTML." That brings me to the wonderful counter argument to the above, which I spend my time running into a lot: "Yes, but that's niche." It's a killer argument, i.e., it kills all discussions completely in one fell swoop. Kind of when you're talking about someone and then this sentence drops into the conversation: "Yes, but she's got cancer now." Here's one implementation of "Yes, but that's niche": Person A: All applications are moving to the web, tablet, and mobile phone. That's especially true now with HTML5, which is going to wipe away everything everywhere and all applications are going to be browser based. Person B: What about air traffic control applications? Will they run on mobile phones too? And do you see defence applications running in a browser? Don't you agree that there are multiple scenarios imaginable where the Java desktop is the optimal platform for running applications? Person A: Yes, but that's niche. Here's another implementation, though it contradicts the above [despite often being used by the same people], since JavaFX is a Java desktop technology: Person A: Swing is dead. Everyone is going to be using purely JavaFX and nothing else. Person B: Does JavaFX have a docking framework and a module system? Does it have a plugin system?  These are some of the absolutely basic requirements of Java desktop software once you get to high end systems, e.g., banks, defence force, oil/gas services. Those kinds of applications need a web browser and so they love the JavaFX WebView component and they also love the animated JavaFX charting components. But they need so much more than that, i.e., an application framework. Aren't there requirements that JavaFX isn't meeting since it is a UI toolkit, just like Swing is a UI toolkit, and what they have in common is their lack, i.e., natively, of any kind of application framework? Don't people need more than a single window and a monolithic application structure? Person A: Yes, but that's niche. In other words, anything that doesn't fit within the currently dominant philosophy is "niche", for no other reason than that it doesn't fit within the currently dominant philosophy... regardless of the actual needs of real developers. Saying "Yes, but that's niche", kills the discussion completely, because it relegates one side of the conversation to the arcane and irrelevant corners of the universe. You're kind of like Cobol now, as soon as "Yes, but that's niche" is said. What's worst about "Yes, but that's niche" is that it doesn't enter into any discussion about user requirements, i.e., there's so few that need this particular solution that we don't even need to talk about them anymore. Note, of course, that I'm not referring specifically or generically to anyone or anything in particular. Just picking up from conversations I've picked up on as I was scurrying around the Hilton's corridors while looking for the location of my next presentation over the past few days. It does, however, mean that there were people thinking "Yes, but that's niche" while listening to James Gosling pointing out that HTML is not the be-all and end-all of absolutely everything. And so this all leaves me wondering: How many applications must be part of a niche for the niche to no longer be a niche? And what if there are multiple small niches that have the same requirements? Don't all those small niches together form a larger whole, one that should be taken seriously, i.e., a whole that is not a niche?

  Read the article

• 12c - SQL Text Expansion

  - by noreply(at)blogger.com (Thomas Kyte)

  Here is another small but very useful new feature in Oracle Database 12c - SQL Text Expansion.  It will come in handy in two cases:You are asked to tune what looks like a simple query - maybe a two table join with simple predicates.  But it turns out the two tables are each views of views of views and so on... In other words, you've been asked to 'tune' a 15 page query, not a two liner.You are asked to take a look at a query against tables with VPD (virtual private database) policies.  In order words, you have no idea what you are trying to 'tune'.A new function, EXPAND_SQL_TEXT, in the DBMS_UTILITY package makes seeing what the "real" SQL is quite easy. For example - take the common view ALL_USERS - we can now:ops$tkyte%ORA12CR1> variable x clobops$tkyte%ORA12CR1> begin  2          dbms_utility.expand_sql_text  3          ( input_sql_text => 'select * from all_users',  4            output_sql_text => :x );  5  end;  6  /PL/SQL procedure successfully completed.ops$tkyte%ORA12CR1> print xX--------------------------------------------------------------------------------SELECT "A1"."USERNAME" "USERNAME","A1"."USER_ID" "USER_ID","A1"."CREATED" "CREATED","A1"."COMMON" "COMMON" FROM  (SELECT "A4"."NAME" "USERNAME","A4"."USER#" "USER_ID","A4"."CTIME" "CREATED",DECODE(BITAND("A4"."SPARE1",128),128,'YES','NO') "COMMON" FROM "SYS"."USER$" "A4","SYS"."TS$" "A3","SYS"."TS$" "A2" WHERE "A4"."DATATS#"="A3"."TS#" AND "A4"."TEMPTS#"="A2"."TS#" AND "A4"."TYPE#"=1) "A1"Now it is easy to see what query is really being executed at runtime - regardless of how many views of views you might have.  You can see the expanded text - and that will probably lead you to the conclusion that maybe that 27 table join to 25 tables you don't even care about might better be written as a two table join.Further, if you've ever tried to figure out what a VPD policy might be doing to your SQL, you know it was hard to do at best.  Christian Antognini wrote up a way to sort of see it - but you never get to see the entire SQL statement: http://www.antognini.ch/2010/02/tracing-vpd-predicates/.  But now with this function - it becomes rather trivial to see the expanded SQL - after the VPD has been applied.  We can see this by setting up a small table with a VPD policy ops$tkyte%ORA12CR1> create table my_table  2  (  data        varchar2(30),  3     OWNER       varchar2(30) default USER  4  )  5  /Table created.ops$tkyte%ORA12CR1> create or replace  2  function my_security_function( p_schema in varchar2,  3                                 p_object in varchar2 )  4  return varchar2  5  as  6  begin  7     return 'owner = USER';  8  end;  9  /Function created.ops$tkyte%ORA12CR1> begin  2     dbms_rls.add_policy  3     ( object_schema   => user,  4       object_name     => 'MY_TABLE',  5       policy_name     => 'MY_POLICY',  6       function_schema => user,  7       policy_function => 'My_Security_Function',  8       statement_types => 'select, insert, update, delete' ,  9       update_check    => TRUE ); 10  end; 11  /PL/SQL procedure successfully completed.And then expanding a query against it:ops$tkyte%ORA12CR1> begin  2          dbms_utility.expand_sql_text  3          ( input_sql_text => 'select * from my_table',  4            output_sql_text => :x );  5  end;  6  /PL/SQL procedure successfully completed.ops$tkyte%ORA12CR1> print xX--------------------------------------------------------------------------------SELECT "A1"."DATA" "DATA","A1"."OWNER" "OWNER" FROM  (SELECT "A2"."DATA" "DATA","A2"."OWNER" "OWNER" FROM "OPS$TKYTE"."MY_TABLE" "A2" WHERE "A2"."OWNER"=USER@!) "A1"Not an earth shattering new feature - but extremely useful in certain cases.  I know I'll be using it when someone asks me to look at a query that looks simple but has a twenty page plan associated with it!

  Read the article

• CPU Usage in Very Large Coherence Clusters

  - by jpurdy

  When sizing Coherence installations, one of the complicating factors is that these installations (by their very nature) tend to be application-specific, with some being large, memory-intensive caches, with others acting as I/O-intensive transaction-processing platforms, and still others performing CPU-intensive calculations across the data grid. Regardless of the primary resource requirements, Coherence sizing calculations are inherently empirical, in that there are so many permutations that a simple spreadsheet approach to sizing is rarely optimal (though it can provide a good starting estimate). So we typically recommend measuring actual resource usage (primarily CPU cycles, network bandwidth and memory) at a given load, and then extrapolating from those measurements. Of course there may be multiple types of load, and these may have varying degrees of correlation -- for example, an increased request rate may drive up the number of objects "pinned" in memory at any point, but the increase may be less than linear if those objects are naturally shared by concurrent requests. But for most reasonably-designed applications, a linear resource model will be reasonably accurate for most levels of scale. However, at extreme scale, sizing becomes a bit more complicated as certain cluster management operations -- while very infrequent -- become increasingly critical. This is because certain operations do not naturally tend to scale out. In a small cluster, sizing is primarily driven by the request rate, required cache size, or other application-driven metrics. In larger clusters (e.g. those with hundreds of cluster members), certain infrastructure tasks become intensive, in particular those related to members joining and leaving the cluster, such as introducing new cluster members to the rest of the cluster, or publishing the location of partitions during rebalancing. These tasks have a strong tendency to require all updates to be routed via a single member for the sake of cluster stability and data integrity. Fortunately that member is dynamically assigned in Coherence, so it is not a single point of failure, but it may still become a single point of bottleneck (until the cluster finishes its reconfiguration, at which point this member will have a similar load to the rest of the members). The most common cause of scaling issues in large clusters is disabling multicast (by configuring well-known addresses, aka WKA). This obviously impacts network usage, but it also has a large impact on CPU usage, primarily since the senior member must directly communicate certain messages with every other cluster member, and this communication requires significant CPU time. In particular, the need to notify the rest of the cluster about membership changes and corresponding partition reassignments adds stress to the senior member. Given that portions of the network stack may tend to be single-threaded (both in Coherence and the underlying OS), this may be even more problematic on servers with poor single-threaded performance. As a result of this, some extremely large clusters may be configured with a smaller number of partitions than ideal. This results in the size of each partition being increased. When a cache server fails, the other servers will use their fractional backups to recover the state of that server (and take over responsibility for their backed-up portion of that state). The finest granularity of this recovery is a single partition, and the single service thread can not accept new requests during this recovery. Ordinarily, recovery is practically instantaneous (it is roughly equivalent to the time required to iterate over a set of backup backing map entries and move them to the primary backing map in the same JVM). But certain factors can increase this duration drastically (to several seconds): large partitions, sufficiently slow single-threaded CPU performance, many or expensive indexes to rebuild, etc. The solution of course is to mitigate each of those factors but in many cases this may be challenging. Larger clusters also lead to the temptation to place more load on the available hardware resources, spreading CPU resources thin. As an example, while we've long been aware of how garbage collection can cause significant pauses, it usually isn't viewed as a major consumer of CPU (in terms of overall system throughput). Typically, the use of a concurrent collector allows greater responsiveness by minimizing pause times, at the cost of reducing system throughput. However, at a recent engagement, we were forced to turn off the concurrent collector and use a traditional parallel "stop the world" collector to reduce CPU usage to an acceptable level. In summary, there are some less obvious factors that may result in excessive CPU consumption in a larger cluster, so it is even more critical to test at full scale, even though allocating sufficient hardware may often be much more difficult for these large clusters.

  Read the article

• OS8- AK8- The bad news...

  - by Steve Tunstall

  Ok I told you I would give you the bad news of AK8 to go along with all the cool new stuff, so here it is. It's not that bad, really, just things you need to be aware of. First, the 2013.1 code is being called OS8, AK8 and 2013.1 by different people. I mean different people INSIDE Oracle!! It was supposed to be easy, but it never is. So for the rest of this blog entry, I'm calling it AK8. AK8 is not compatible with the 7x10 series. Ever. The 7x10 series is not supported with AK8, and if you try to upgrade one, it will fail at the healthcheck. All 7x20 series, all of them regardless of age, are supported with AK8. Drive trays. Let's talk about drive trays and SAS cards. The older drive trays for the 7x20 series were called the "Riverwalk 2" or "DS2" trays. They were technically the "J4410" series JBODs that Sun used to sell a la carte before we stopped selling JBODs. Don't get me started on that, it still makes me mad. We used these for many years, and you can still buy them right now until December 15th, 2013, when they will no longer be sold. The DS2 tray only came as a 4u, 24 drive shelf. It held 3.5" drives, and you had a choice of 2TB, 3TB, 300GB or 600GB drives. The SAS HBA in the 7x20 series was called a "Thebe" card, with a part # of 7105394. The 7420, for example, came standard with two of these "Thebe" cards for connecting to the disk trays. Two Thebe cards could handle up to 12 trays, so one would add two more cards to go to 24 trays, or have up to six Thebe cards to handle 36 trays. This card was for external SAS only. It did not connect to the internal OS drives or the Readzillas, both of which used the internal SCSI controller of the server. These Riverwalk 2 trays ARE supported with AK8. You can upgrade your older 7420 or 7320, no problem, as-is. The much older Riverwalk 1 trays or J4400 trays are NOT supported by AK8. However, they were only used by the 7x10 series, and we already said that the 7x10 series was not supported. Here's where it gets tricky. Since last January, we have been selling the new style disk trays. We call them the "DE2-24P" and the "DE2-24C" trays. The "C" tray is for capacity drives, which are 3.5" 3TB or 4TB drives. The "P" trays are for performance drives, which are 2.5" 300GB and 900GB drives. These trays are NOT Riverwalk 2 trays, even though the "C" series may kind of look like it. Different manufacturer and different firmware. They are not new. Like I said, we've been selling them with the 7x20 series since last January. They are the only disk trays we will be selling going forward. Of course, AK8 supports them. So what's the problem? The problem is going to be for people who have to mix drive trays. Remember, your older 7x20 series has Thebe SAS2 HBAs. These have 2 SAS ports per card.  The new ZS3-2 and ZS3-4 systems, however, have the new "Thebe2" SAS2 HBAs. These Thebe2 cards have 4 ports per card. This is very cool, as we can now do more SAS channels with less cards. Instead of needing 4 SAS cards to grow to 24 trays like we did with the old Thebe cards, I can now do 24 trays with only 2 Thebe2 cards. This means more IO slots for fun things like Infiniband and 10G. So far, so good, right? These Thebe2 cards work with any disk tray. You can even mix older DS2 trays with the newer DE2 trays in the same system, as long as you have Thebe2 cards. Ah, there's your problem. You don't have Thebe2 cards in your old 7420, do you? Well, I told you the bad news wasn't that bad, right? We can take out your Thebe cards and replace them with Thebe2. You can then plug your older DS2 trays right back in, and also now get newer DE2 trays going forward. However, it's important that the trays are on different SAS channels. You can mix them in the same system, but not on the same channel. Ask your local SC if you need help with the new cable layout. By the way, the new ZS3-2 and ZS3-4 systems also include a new IO card called "Erie" cards. These are for INTERNAL SAS to the OS drives and the Readzillas. So those are now SAS2 instead of SATA like the older models. Yes, the Erie card uses an IO slot, but that's OK, because the Thebe2 cards allow us to use less SAS HBAs to grow the system, right? That's it. Not too much bad news and really not that bad. AK8 does not support the 7x10 series, and you may need new Thebe2 cards in your older systems if you want to add on newer DE2 trays. I think we can all agree that there are worse things out there. Like our Congress.   Next up.... More good news and cool AK8 tricks. Such as virtual NICS. 

  Read the article

• VNIC - New feature of AK8 - Working with VNICs

  - by Steve Tunstall

  One of the important new features of the AK8 code is the ability to use multiple IP addresses on the same physical network port. This feature is called VNICs, or Virtual NICs. This allows us to no longer "burn" a whole port in a cluster when one cluster peer owns a network port. Traditionally, we have had to leave Net0 empty on controller 2, because it was used for managing controller 1. Vise-versa for Net1 on Controller 1. Then, if you have data going over 10GigE ports, you probably only had half of your ports running at any given time, and the partner 10GigE port on the other controller just sat there, doing nothing, unless the first controller went down. What a waste. Those days are over.  I want to thank and give a big shout-out to our good partner, OnX Enterprise Solutions, for allowing me to come into their lab and play around with their 7320 to do this demo. They let me make a big mess of their lab for the day as I played around with VNICs. If you're looking for a partner who knows Oracle well and can also piece together a solution from multiple vendors to get you what you need, OnX is a good choice. If you would like to talk to your local OnX rep, you can contact Scott Gill at [email protected] and he can point you in the right direction for your area.  Here we go: Here is what your Datalinks window looks like BEFORE you upgrade to AK8. Here's what the same screen looks like after you upgrade. See the new box? So here is my current network setup. I have my 4 physical interfaces setup each with an IP address. If I ping them, no problems.  So I can ping 180, 181, 251, and 252. However, if I try to ping 240, it does not work, as the 240 address is not being used by any of these interfaces, right?Let's change that. Here, I'm going to make a new Datalink by clicking the Datalink "Plus sign" button. I will check the VNIC box and tell it to use igb2, even though another interface is already using it. Now, I will create a new Interface, and choose "v_dl2" for it's datalink. My new network screen looks like this. A few things to take note of here. First, when I click the "igb2" device, it only highlights dl2 and int2. It does not highlight v_dl2 or v_int2.I think it should, but OK, it looks like VNICs don't highlight when you click the device. Second, note how the underscore character in v_dl2 and v_int2 do not seem to show on this screen. You can see it plainly if you go in and edit them, but from here it looks like a space instead of an underscore. Just a cosmetic bug, but something to be aware of. Now, if I click the VNIC datalink "v_dl2", on the other hand, it DOES highlight the device it belongs to, as it should. Seen here: Note that it did not, however, highlight int2 with it, even though int2 is connected to igb2. That's because we clicked v_dl2, which int2 has nothing to do with. So I'm OK with that. So let's try pinging 240 now. Of course, it works great.  So I now make another VNIC, and call it v_dl3 using igb3, and v_int3 with an address of 241. I then setup three shares, using ports 251, 240, and 241.Remember that IP 251 and 240 both are using the same physical port of igb2, and IP 241 is using port igb3. Next, I copy a folder full of stuff over to all three shares at the same time. I have analytics going so I can see the traffic. My top chart is showing the logical interfaces, and the bottom chart is showing the physical ports.Sure enough, look at the igb2 and vnic1 interfaces. They equal the traffic going over the igb2 physical port on the second chart. VNIC2, on the other hand, gets igb3 all to itself. This would work the same way with 10Gig or Infiniband ports. You can now have multiple IP addresses and even completely different subnets sharing the same physical ports. You may need to make route table entries for that. This allows us to use all of the ports you paid for with no more waste.  Very, very cool.  One small "bug" I found when doing this. It's really not a bug, it was designed to do this when VNICs were not around. But now that we have NVIC capability, they should probably change this. I've alerted the engineering team about this and they're looking into it, so perhaps it will be fixed in a later code. Here it is. Remember when we made the new VNIC datalink, I specifically said to click on the "Plus Sign" button to create it? I don't always do that. I really like to use the drag-and-drop method to create my datalinks in the network screen.HOWEVER, if you were to do that for building a VNIC, it will mess you up a little. Watch this. Here, I'm dragging igb3 over to make a new datalink. igb3 is already being used by dl3, but I'm going to make this a VNIC, so who cares, right? Well, the ZFSSA does not KNOW you are going to make it a VNIC, now does it? So... it works as designed and REMOVES the igb3 device from the current dl3 datalink in the background. See how it's now missing? At the same time, the dl3 datalink choice is missing from my list of possible VNICs for me to choose from!!!! Hey!!! I wanted to pick dl3. Why isn't it on the list??? Well, it can't be on this list because dl3 no longer has a device associated with it. Bummer for you. When you click cancel, the device is still missing from dl3. The fix is easy. Just edit dl3 by clicking the pencil button, do absolutely nothing, and click "Apply". The device will magically come back. Now, make the VNIC datalink by clicking the "Plus Sign" button. Sure enough, once you check the VNIC box, dl3 is a valid choice. No problem.  That's it for now. Have fun with VNICs.

  Read the article

• How to Plug a Small Hole in NetBeans JSF (Join Table) Code Generation

  - by MarkH

  I was asked recently to provide an assist with designing and building a small-but-vital application that had at its heart some basic CRUD (Create, Read, Update, & Delete) functionality, built upon an Oracle database, to be accessible from various locations. Working from the stated requirements, I fleshed out the basic application and database designs and, once validated, set out to complete the first iteration for review. Using SQL Developer, I created the requisite tables, indices, and sequences for our first run. One of the tables was a many-to-many join table with three fields: one a primary key for that table, the other two being primary keys for the other tables, represented as foreign keys in the join table. Here is a simplified example of the trio of tables: Once the database was in decent shape, I fired up NetBeans to let it have first shot at the code. NetBeans does a great job of generating a mountain of essential code, saving developers what must be millions of hours of effort each year by building a basic foundation with a few clicks and keystrokes. Lest you think it (or any tool) can do everything for you, however, occasionally something tosses a paper clip into the delicate machinery and makes you open things up to fix them. Join tables apparently qualify.  :-) In the case above, the entity class generated for the join table (New Entity Classes from Database) included an embedded object consisting solely of the two foreign key fields as attributes, in addition to an object referencing each one of the "component" tables. The Create page generated (New JSF Pages from Entity Classes) worked well to a point, but when trying to save, we were greeted with an error: Transaction aborted. Hmm. A quick debugger session later and I'd identified the issue: when trying to persist the new join-table object, the embedded "foreign-keys-only" object still had null values for its two (required value) attributes...even though the embedded table objects had populated key attributes. Here's the simple fix: In the join-table controller class, find the public String create() method. It will look something like this:     public String create() {        try {            getFacade().create(current);            JsfUtil.addSuccessMessage(ResourceBundle.getBundle("/Bundle").getString("JoinEntityCreated"));            return prepareCreate();        } catch (Exception e) {            JsfUtil.addErrorMessage(e, ResourceBundle.getBundle("/Bundle").getString("PersistenceErrorOccured"));            return null;        }    } To restore balance to the force, modify the create() method as follows (changes in red):     public String create() {         try {            // Add the next two lines to resolve:            current.getJoinEntityPK().setTbl1id(current.getTbl1().getId().toBigInteger());            current.getJoinEntityPK().setTbl2id(current.getTbl2().getId().toBigInteger());            getFacade().create(current);            JsfUtil.addSuccessMessage(ResourceBundle.getBundle("/Bundle").getString("JoinEntityCreated"));            return prepareCreate();        } catch (Exception e) {            JsfUtil.addErrorMessage(e, ResourceBundle.getBundle("/Bundle").getString("PersistenceErrorOccured"));            return null;        }    } I'll be refactoring this code shortly, but for now, it works. Iteration one is complete and being reviewed, and we've met the milestone. Here's to happy endings (and customers)! All the best,Mark

  Read the article

• Coherence Data Guarantees for Data Reads - Basic Terminology

  - by jpurdy

  When integrating Coherence into applications, each application has its own set of requirements with respect to data integrity guarantees. Developers often describe these requirements using expressions like "avoiding dirty reads" or "making sure that updates are transactional", but we often find that even in a small group of people, there may be a wide range of opinions as to what these terms mean. This may simply be due to a lack of familiarity, but given that Coherence sits at an intersection of several (mostly) unrelated fields, it may be a matter of conflicting vocabularies (e.g. "consistency" is similar but different in transaction processing versus multi-threaded programming). Since almost all data read consistency issues are related to the concept of concurrency, it is helpful to start with a definition of that, or rather what it means for two operations to be concurrent. Rather than implying that they occur "at the same time", concurrency is a slightly weaker statement -- it simply means that it can't be proven that one event precedes (or follows) the other. As an example, in a Coherence application, if two client members mutate two different cache entries sitting on two different cache servers at roughly the same time, it is likely that one update will precede the other by a significant amount of time (say 0.1ms). However, since there is no guarantee that all four members have their clocks perfectly synchronized, and there is no way to precisely measure the time it takes to send a given message between any two members (that have differing clocks), we consider these to be concurrent operations since we can not (easily) prove otherwise. So this leads to a question that we hear quite frequently: "Are the contents of the near cache always synchronized with the underlying distributed cache?". It's easy to see that if an update on a cache server results in a message being sent to each near cache, and then that near cache being updated that there is a window where the contents are different. However, this is irrelevant, since even if the application reads directly from the distributed cache, another thread update the cache before the read is returned to the application. Even if no other member modifies a cache entry prior to the local near cache entry being updated (and subsequently read), the purpose of reading a cache entry is to do something with the result, usually either displaying for consumption by a human, or by updating the entry based on the current state of the entry. In the former case, it's clear that if the data is updated faster than a human can perceive, then there is no problem (and in many cases this can be relaxed even further). For the latter case, the application must assume that the value might potentially be updated before it has a chance to update it. This almost aways the case with read-only caches, and the solution is the traditional optimistic transaction pattern, which requires the application to explicitly state what assumptions it made about the old value of the cache entry. If the application doesn't want to bother stating those assumptions, it is free to lock the cache entry prior to reading it, ensuring that no other threads will mutate the entry, a pessimistic approach. The optimistic approach relies on what is sometimes called a "fuzzy read". In other words, the application assumes that the read should be correct, but it also acknowledges that it might not be. (I use the qualifier "sometimes" because in some writings, "fuzzy read" indicates the situation where the application actually sees an original value and then later sees an updated value within the same transaction -- however, both definitions are roughly equivalent from an application design perspective). If the read is not correct it is called a "stale read". Going back to the definition of concurrency, it may seem difficult to precisely define a stale read, but the practical way of detecting a stale read is that is will cause the encompassing transaction to roll back if it tries to update that value. The pessimistic approach relies on a "coherent read", a guarantee that the value returned is not only the same as the primary copy of that value, but also that it will remain that way. In most cases this can be used interchangeably with "repeatable read" (though that term has additional implications when used in the context of a database system). In none of cases above is it possible for the application to perform a "dirty read". A dirty read occurs when the application reads a piece of data that was never committed. In practice the only way this can occur is with multi-phase updates such as transactions, where a value may be temporarily update but then withdrawn when a transaction is rolled back. If another thread sees that value prior to the rollback, it is a dirty read. If an application uses optimistic transactions, dirty reads will merely result in a lack of forward progress (this is actually one of the main risks of dirty reads -- they can be chained and potentially cause cascading rollbacks). The concepts of dirty reads, fuzzy reads, stale reads and coherent reads are able to describe the vast majority of requirements that we see in the field. However, the important thing is to define the terms used to define requirements. A quick web search for each of the terms in this article will show multiple meanings, so I've selected what are generally the most common variations, but it never hurts to state each definition explicitly if they are critical to the success of a project (many applications have sufficiently loose requirements that precise terminology can be avoided).

  Read the article

• Implementing a Custom Coherence PartitionAssignmentStrategy

  - by jpurdy

  A recent A-Team engagement required the development of a custom PartitionAssignmentStrategy (PAS). By way of background, a PAS is an implementation of a Java interface that controls how a Coherence partitioned cache service assigns partitions (primary and backup copies) across the available set of storage-enabled members. While seemingly straightforward, this is actually a very difficult problem to solve. Traditionally, Coherence used a distributed algorithm spread across the cache servers (and as of Coherence 3.7, this is still the default implementation). With the introduction of the PAS interface, the model of operation was changed so that the logic would run solely in the cache service senior member. Obviously, this makes the development of a custom PAS vastly less complex, and in practice does not introduce a significant single point of failure/bottleneck. Note that Coherence ships with a default PAS implementation but it is not used by default. Further, custom PAS implementations are uncommon (this engagement was the first custom implementation that we know of). The particular implementation mentioned above also faced challenges related to managing multiple backup copies but that won't be discussed here. There were a few challenges that arose during design and implementation: Naive algorithms had an unreasonable upper bound of computational cost. There was significant complexity associated with configurations where the member count varied significantly between physical machines. Most of the complexity of a PAS is related to rebalancing, not initial assignment (which is usually fairly simple). A custom PAS may need to solve several problems simultaneously, such as: Ensuring that each member has a similar number of primary and backup partitions (e.g. each member has the same number of primary and backup partitions) Ensuring that each member carries similar responsibility (e.g. the most heavily loaded member has no more than one partition more than the least loaded). Ensuring that each partition is on the same member as a corresponding local resource (e.g. for applications that use partitioning across message queues, to ensure that each partition is collocated with its corresponding message queue). Ensuring that a given member holds no more than a given number of partitions (e.g. no member has more than 10 partitions) Ensuring that backups are placed far enough away from the primaries (e.g. on a different physical machine or a different blade enclosure) Achieving the above goals while ensuring that partition movement is minimized. These objectives can be even more complicated when the topology of the cluster is irregular. For example, if multiple cluster members may exist on each physical machine, then clearly the possibility exists that at certain points (e.g. following a member failure), the number of members on each machine may vary, in certain cases significantly so. Consider the case where there are three physical machines, with 3, 3 and 9 members each (respectively). This introduces complexity since the backups for the 9 members on the the largest machine must be spread across the other 6 members (to ensure placement on different physical machines), preventing an even distribution. For any given problem like this, there are usually reasonable compromises available, but the key point is that objectives may conflict under extreme (but not at all unlikely) circumstances. The most obvious general purpose partition assignment algorithm (possibly the only general purpose one) is to define a scoring function for a given mapping of partitions to members, and then apply that function to each possible permutation, selecting the most optimal permutation. This would result in N! (factorial) evaluations of the scoring function. This is clearly impractical for all but the smallest values of N (e.g. a partition count in the single digits). It's difficult to prove that more efficient general purpose algorithms don't exist, but the key take away from this is that algorithms will tend to either have exorbitant worst case performance or may fail to find optimal solutions (or both) -- it is very important to be able to show that worst case performance is acceptable. This quickly leads to the conclusion that the problem must be further constrained, perhaps by limiting functionality or by using domain-specific optimizations. Unfortunately, it can be very difficult to design these more focused algorithms. In the specific case mentioned, we constrained the solution space to very small clusters (in terms of machine count) with small partition counts and supported exactly two backup copies, and accepted the fact that partition movement could potentially be significant (preferring to solve that issue through brute force). We then used the out-of-the-box PAS implementation as a fallback, delegating to it for configurations that were not supported by our algorithm. Our experience was that the PAS interface is quite usable, but there are intrinsic challenges to designing PAS implementations that should be very carefully evaluated before committing to that approach.

  Read the article

• Synchronized Property Changes (Part 4)

  - by Geertjan

  The next step is to activate the undo/redo functionality... for a Node. Something I've not seen done before. I.e., when the Node is renamed via F2 on the Node, the "Undo/Redo" buttons should start working. Here is the start of the solution, via this item in the mailing list and Timon Veenstra's BeanNode class, note especially the items in bold: public class ShipNode extends BeanNode implements PropertyChangeListener, UndoRedo.Provider { private final InstanceContent ic; private final ShipSaveCapability saveCookie; private UndoRedo.Manager manager; private String oldDisplayName; private String newDisplayName; private Ship ship; public ShipNode(Ship bean) throws IntrospectionException { this(bean, new InstanceContent()); } private ShipNode(Ship bean, InstanceContent ic) throws IntrospectionException { super(bean, Children.LEAF, new ProxyLookup(new AbstractLookup(ic), Lookups.singleton(bean))); this.ic = ic; setDisplayName(bean.getType()); setShortDescription(String.valueOf(bean.getYear())); saveCookie = new ShipSaveCapability(bean); bean.addPropertyChangeListener(WeakListeners.propertyChange(this, bean)); } @Override public Action[] getActions(boolean context) { List<? extends Action> shipActions = Utilities.actionsForPath("Actions/Ship"); return shipActions.toArray(new Action[shipActions.size()]); } protected void fire(boolean modified) { if (modified) { ic.add(saveCookie); } else { ic.remove(saveCookie); } } @Override public UndoRedo getUndoRedo() { manager = Lookup.getDefault().lookup( UndoRedo.Manager.class); return manager; } private class ShipSaveCapability implements SaveCookie { private final Ship bean; public ShipSaveCapability(Ship bean) { this.bean = bean; } @Override public void save() throws IOException { StatusDisplayer.getDefault().setStatusText("Saving..."); fire(false); } } @Override public boolean canRename() { return true; } @Override public void setName(String newDisplayName) { Ship c = getLookup().lookup(Ship.class); oldDisplayName = c.getType(); c.setType(newDisplayName); fireNameChange(oldDisplayName, newDisplayName); fire(true); fireUndoableEvent("type", ship, oldDisplayName, newDisplayName); } public void fireUndoableEvent(String property, Ship source, Object oldValue, Object newValue) { ReUndoableEdit reUndoableEdit = new ReUndoableEdit( property, source, oldValue, newValue); UndoableEditEvent undoableEditEvent = new UndoableEditEvent( this, reUndoableEdit); manager.undoableEditHappened(undoableEditEvent); } private class ReUndoableEdit extends AbstractUndoableEdit { private Object oldValue; private Object newValue; private Ship source; private String property; public ReUndoableEdit(String property, Ship source, Object oldValue, Object newValue) { super(); this.oldValue = oldValue; this.newValue = newValue; this.source = source; this.property = property; } @Override public void undo() throws CannotUndoException { setName(oldValue.toString()); } @Override public void redo() throws CannotRedoException { setName(newValue.toString()); } } @Override public String getDisplayName() { Ship c = getLookup().lookup(Ship.class); if (null != c.getType()) { return c.getType(); } return super.getDisplayName(); } @Override public String getShortDescription() { Ship c = getLookup().lookup(Ship.class); if (null != String.valueOf(c.getYear())) { return String.valueOf(c.getYear()); } return super.getShortDescription(); } @Override public void propertyChange(PropertyChangeEvent evt) { if (evt.getPropertyName().equals("type")) { String oldDisplayName = evt.getOldValue().toString(); String newDisplayName = evt.getNewValue().toString(); fireDisplayNameChange(oldDisplayName, newDisplayName); } else if (evt.getPropertyName().equals("year")) { String oldToolTip = evt.getOldValue().toString(); String newToolTip = evt.getNewValue().toString(); fireShortDescriptionChange(oldToolTip, newToolTip); } fire(true); } } Undo works when rename is done, but Redo never does, because Undo is constantly activated, since it is reactivated whenever there is a name change. And why must the UndoRedoManager be retrieved from the Lookup (it doesn't work otherwise)? Don't get that part of the code either. Help welcome!

  Read the article

• Twitter ?? Nashorn ????(??)

  - by Homma

  ???? Nashorn ? Java ??????? Twitter ???????????????????? JavaFX ??????????????? ????? ??? jlaskey ??? Nashorn Blog ????????????? https://blogs.oracle.com/nashorn/entry/nashorn_in_the_twitterverse_continued ???????? ?? Twitter ???????????????????????? JavaFX ??????????????????????????????? Nashorn ?? JavaFX ??????????????JavaFX ???????????????????????????????????????Nashorn ? Java ????????????????????????????????????(JavaFX ?????????????????????)? ?????????????????????????????????????????????? Twitter ????????????????????????? var twitter4j = Packages.twitter4j; var TwitterFactory = twitter4j.TwitterFactory; var Query = twitter4j.Query; function getTrendingData() { var twitter = new TwitterFactory().instance; var query = new Query("nashorn OR nashornjs"); query.since("2012-11-21"); query.count = 100; var data = {}; do { var result = twitter.search(query); var tweets = result.tweets; for each (var tweet in tweets) { var date = tweet.createdAt; var key = (1900 + date.year) + "/" + (1 + date.month) + "/" + date.date; data[key] = (data[key] || 0) + 1; } } while (query = result.nextQuery()); return data; } ??????????????????getTrendingData() ??????????????(??????????Nashorn ???????? OpenJDK ?????? 2012 ? 11 ? 21 ???)??????????????????????????????????? ????JavaFX ? BarChart ??????????? var javafx = Packages.javafx; var Stage = javafx.stage.Stage var Scene = javafx.scene.Scene; var Group = javafx.scene.Group; var Chart = javafx.scene.chart.Chart; var FXCollections = javafx.collections.FXCollections; var ObservableList = javafx.collections.ObservableList; var CategoryAxis = javafx.scene.chart.CategoryAxis; var NumberAxis = javafx.scene.chart.NumberAxis; var BarChart = javafx.scene.chart.BarChart; var XYChart = javafx.scene.chart.XYChart; var Series = javafx.scene.chart.XYChart.Series; var Data = javafx.scene.chart.XYChart.Data; function graph(stage, data) { var root = new Group(); stage.scene = new Scene(root); var dates = Object.keys(data); var xAxis = new CategoryAxis(); xAxis.categories = FXCollections.observableArrayList(dates); var yAxis = new NumberAxis("Tweets", 0.0, 200.0, 50.0); var series = FXCollections.observableArrayList(); for (var date in data) { series.add(new Data(date, data[date])); } var tweets = new Series("Tweets", series); var barChartData = FXCollections.observableArrayList(tweets); var chart = new BarChart(xAxis, yAxis, barChartData, 25.0); root.children.add(chart); } ????????????????????????????????stage.scene = new Scene(root) ? stage.setScene(new Scene(root)) ????????????????????Nashorn ? stage ??????? scene ???????????????????(Dynalink ?????????)Java Beans ???????????????? (setScene()) ???????????????????????????????Nashorn ? FXCollections ??????????????????????????????observableArrayList(dates) ??????????Nashorn ? JavaScript ??? (dates) ? Java ???????????????????????????? JavaScript ?????????????????? Java ????????????????????????????????????????????????????????????? ????????????????????????????????? JavaFX ???????????????????????? JavaFX ??????????????javafx.application.Application ??????????????????????????? JavaFX ????????????????????????????????????????????????? import java.io.IOException; import java.io.InputStream; import java.io.InputStreamReader; import javafx.application.Application; import javafx.stage.Stage; import javax.script.ScriptEngine; import javax.script.ScriptEngineManager; import javax.script.ScriptException; public class TrendingMain extends Application { private static final ScriptEngineManager MANAGER = new ScriptEngineManager(); private final ScriptEngine engine = MANAGER.getEngineByName("nashorn"); private Trending trending; public static void main(String[] args) { launch(args); } @Override public void start(Stage stage) throws Exception { trending = (Trending) load("Trending.js"); trending.start(stage); } @Override public void stop() throws Exception { trending.stop(); } private Object load(String script) throws IOException, ScriptException { try (final InputStream is = TrendingMain.class.getResourceAsStream(script)) { return engine.eval(new InputStreamReader(is, "utf-8")); } } } ???? Nashorn ??????? JSR-223 ? javax.script ????????? private static final ScriptEngineManager MANAGER = new ScriptEngineManager(); private final ScriptEngine engine = MANAGER.getEngineByName("nashorn"); ????????? JavaScript ???????? Nashorn ???????????????????? load ???????????????????????engine ???????????????load ????????????? ???????????????Java ???????????????????????????????????????????????????? Java ????????????????JavaFX ???????? start ????? stop ?????????????????????????????????????? public interface Trending { public void start(Stage stage) throws Exception; public void stop() throws Exception; } ?????????????????????????????? function newTrending() { return new Packages.Trending() { start: function(stage) { var data = getTrendingData(); graph(stage, data); stage.show(); }, stop: function() { } } } newTrending(); ?????? Trending ?????????????????????start ????? stop ??????????????????????????????????? eval ???? Java ??????????????? trending = (Trending) load("Trending.js"); ????????????????Trending.js ??????? getTrendingData ???????????? newTrending ????????????????????? Java ?????????newTrending ????????? eval ????????? Trending ????????????????????????????????????????????????????????? trending.start(stage); ???????? ???? Nashorn ????????? http://www.myexpospace.com/JavaOne2012/SessionFiles/CON5251_PDF_5251_0001.pdf ???????? Dynalink ??????? https://github.com/szegedi/dynalink ????????

  Read the article

• How to escape or remove double quotes in rsyslog template

  - by Evgeny

  I want rsyslog to write log messages in JSON format, which requires to use double-quotes (") around strings. Problem is that values sometime include double-quotes themselves, and those need to be escaped - but I can't figure out how to do that. Currently my rsyslog.conf contains this format that I use (a bit simplified): $template JsonFormat,"{\"msg\":\"%msg%\",\"app-name\":\"%app-name%\"}\n",sql But when a msg arrives that contains double quotes, the JSON is broken, example: user pid=21214 uid=0 auid=4294967295 msg='PAM setcred: user="oracle" exe="/bin/su" (hostname=?, addr=?, terminal=? result=Success)' turns into: {"msg":"user pid=21214 uid=0 auid=4294967295 msg='PAM setcred: user="oracle" exe="/bin/su" (hostname=?, addr=?, terminal=? result=Success)'","app-name":"user"} but what I need it to become is: {"msg":"user pid=21214 uid=0 auid=4294967295 msg='PAM setcred: user=\"oracle\" exe=\"/bin/su\" (hostname=?, addr=?, terminal=? result=Success)'","app-name":"user"}

  Read the article

• How can I create an “su” only user (no SSH or SFTP) and limit who can “su” into that account in RHEL5? [closed]

  - by Beaming Mel-Bin

  Possible Duplicate: How can I allow one user to su to another without allowing root access? We have a user account that our DBAs use (oracle). I do not want to set a password on this account and want to only allow users in the dba group to su - oracle. How can I accomplish this? I was thinking of just giving them sudo access to the su - oracle command. However, I wouldn't be surprised if there was a more polished/elegant/secure way.

  Read the article

• Difficulty in running Tomcat v7.0 with Eclipse Juno

  - by user1673718

  I get the following error when I run my JSP file in Eclipse-Juno with Tomcat v7: 'starting Tomcat v7.0 server at localhost' has encountered a problem. Port 8080 required by Tomcat v7.0 server at localhost is already in use. The server may already be running in another process, or a system process may be using the port. To start this server you will need to stop the other process or change the port number(s). I have Oracle 10g installed in my System. When I type "http://localhost:8080" it opens the Oracle 10g license agreement so I think Oracle 10g is already running in that port. To change the port of Tomcat I tried Google, which said to change the port in the "C:\Program Files\Apache Software Foundation\Apache Tomcat 7.0.14\conf\httpd.conf" file But at "C:\Program Files\Apache Software Foundation\Apache Tomcat 7.0.14\conf" there was no httpd.conf file. I only have "catalina.policy,catalina.properties,context,logging.properties,server,tomcat-users,web" files in that conf folder. I use windows XP.

  Read the article

• Where is the central ZFS website now?

  - by Stefan Lasiewski

  Oracle dumped OpenSolaris in Fall 2010, and it is unclear if Oracle will continue to publicly release updates to ZFS, except maybe after they release their next major version of Solaris. FreeBSD now has ZFS v28 available for testing. But where did v28 come from? I notice that the main ZFS website does not show version 28 available. Has this website been abandoned? If so, where is the central website for the ZFS project, so that I can browse the repo, read the mailing lists, read the release notes, etc. (I realize that OpenSolaris has been dumped by Oracle, and that they are limiting their ZFS releases to the community).

  Read the article

< Previous Page | 562 563 564 565 566 567 568 569 570 571 572 573  | Next Page >