On 25/02/2013 12:45 AM, Tom Lane wrote:
> Odd. I wonder if you are seeing some variant of the old context swap storm problem. The "99.8% system time" reading
issuggestive but hardly conclusive. Does top's report of context swap rate go to the moon? It would be interesting to
stracea few of the server processes while one of these events is happening, too. regards, tom lane
I used vmstat to look at the context swaps, they were around 5k and 15k interrupts per second.
I thought that it was to many interrupts and after a bit of search a website mentioned that the network card driver
couldcause that. After updating kernel and the driver the stalling is not reproducible any more.
Weird enough, when I load test the server now I have 35k interrupts and 250k context switches, but no problems at all.
Thanks for pointing me into the right direction.
