On 2013-01-22, Илья Шипицин <chipits...@gmail.com> wrote: > I'm running https web reverse proxy. > at 200-500mbit scale, I see 3500 interrupts per second at em0, em1, also 12 > cpus are running at 70-80%, > CPU00 is running at "interrupt" level, also there're user processes at > "user" and "system" levels. > > > under such load server is experience somewhat to "general network delays", > network conections become slow (both incoming and outgoing), sometimes even > 5 sec on 1G network.
There are various other possibilities to look at before you go messing with the scheduler. Slow network connections - could you be running out of free TCP ports (including connections in TIME_WAIT if not SO_REUSEADDR) and so the system has to wait for old connections to close down?
