Flamegraph of 8KB message bottlenecks

[Adrian Chadd]

On 21 January 2016 at 14:34, Matthew Macy <mmacy at nextbsd.org> wrote:
> ~170k interrupts / second for ~800kpps, but at this point clearly all the time is spent on syscall overhead

Sure, that's why I asked.

I see the fgets stuff looking lock contend-y, and I see a sleeplock
kicking in via kern_sendit -> sx_xlock_hard; so something's blocking
on a lock. The kqueue locking too is a problem - kqueue doesn't handle
lots of fast producer/consumer things and it contends badly. :(


-a

>
>
>
>  ---- On Thu, 21 Jan 2016 14:11:34 -0800 Adrian Chadd <adrian.chadd at gmail.com> wrote ----
>  > Well, how many interrupts/second is this seeing?
>  >
>  >
>  > -a
>  > _______________________________________________
>  > freebsd-transport at freebsd.org mailing list
>  > https://lists.freebsd.org/mailman/listinfo/freebsd-transport
>  > To unsubscribe, send any mail to "freebsd-transport-unsubscribe at freebsd.org"
>  >
>