I have a problem building an aggregate function. Here's my problem:

I have a table like this

 id action day         isSum difference
 1  ping   2012-01-01  1     500         (this is the sum of the differences from last year)
 2  ping   2012-01-01  0     -2
 3  ping   2012-01-02  0      1
 4  ping   2012-01-03  0     -4
 5  ping   2012-01-04  0     -2
 6  ping   2012-01-05  0      3
 7  ping   2012-01-06  0      2
 8  ping   2012-01-01  1      0         (this is the sum of the differences from last year, now for pong)
 9  pong   2012-01-01  0     -5
 10  pong   2012-01-02  0      2
 11  pong   2012-01-03  0     -2
 12  pong   2012-01-04  0     -8
 13  pong   2012-01-05  0      3
 14  pong   2012-01-06  0      4

I now need to select the action, day and the summarized difference since 01-01 for every day, so that my result looks like this

action day        total      
ping   2012-01-01 498
ping   2012-01-02 499
ping   2012-01-03 495
ping   2012-01-04 493
ping   2012-01-05 496
ping   2012-01-06 498
pong   2012-01-01 - 5
pong   2012-01-02 - 3
pong   2012-01-03 - 5
pong   2012-01-04 -13
pong   2012-01-05 -10
pong   2012-01-06 - 6

How can I do this?

there a are a lot of datasets (~1 million), so the query needs to be pretty cheap. I don't know how the use sum to get daily sums for daily records depending on the action-column.