Date: Fri, 14 Nov 2008 04:46:39 -0800 (PST) From: Won De Erick <won.derick@yahoo.com> To: Ivan Voras <ivoras@freebsd.org>, freebsd-hardware@freebsd.org Cc: freebsd-net@freebsd.org Subject: Re: IRQ31 and IRQ32 on HPDL585 running FreeBSD 7.0 are consuming HIGH CPU usage Message-ID: <555333.76711.qm@web45806.mail.sp1.yahoo.com> References: <305614.76266.qm@web45809.mail.sp1.yahoo.com> <gfjoio$cf3$1@ger.gmane.org>
next in thread | previous in thread | raw e-mail | index | archive | help
> ----- Original Message ----=0A=0A> From: Ivan Voras <ivoras@freebsd.org>=
=0A> To: freebsd-hardware@freebsd.org=0A> Cc: freebsd-net@freebsd.org=0A> S=
ent: Friday, November 14, 2008 7:49:13 PM=0A> Subject: Re: IRQ31 and IRQ32 =
on HPDL585 running FreeBSD 7.0 are consuming HIGH CPU usage=0A> =0A> Won De=
Erick wrote:=0A> =0A> > Another thing, I observed that in the above test, =
the net.isr is enabled by default. When I tried disabling this,=0A> > =0A> =
> # sysctl net.isr.direct=3D0=0A> > net.isr.direct: 1 -> 0=0A> > =0A> > the=
result:=0A> > =0A> > 52 root 1 -68 - 0K 16K WAIT b 64:=
00 42.97% irq32: bce1=0A> > 51 root 1 -68 - 0K 16K WAIT =
a 38:22 12.26% irq31: bce0=0A> > =0A> > The CPU utilizations considerably =
dropped!=0A> =0A> You will probably find a "swi" process that has picked up=
the difference=0A> (when isr.direct is disabled, some of network protocol =
processing is=0A> offloaded to a swi thread). This might help spread the lo=
ad across CPU=0A> but in my testing it didn't help real-world throughput.=
=0A>=0A=0AYou are right. I noticed the following when net.isr is diabled, l=
owering the idle time of cpu0.=0A=0A 27 root 1 -44 - 0K =
16K WAIT 0 52:20 76.37% swi1: net=0A 26 root 1 171 ki31 0K =
16K CPU0 0 111:58 64.36% idle: cpu0=0A=0A=0AAnother thing,=0APacket dr=
ops on Intel NIC ( Intel=AE PRO/1000 PT Dual Port Server Adapter w/ control=
processor 82571GB) did not occur when the net.isr was disabled, while the =
overall CPU utilization remains considerably low.=0A=0ANote: The following =
result was obtained during a transition from a disabled to enabled net.isr.=
Hence the first part=0A=0Apackets errs bytes packets errs b=
ytes colls drops=0A 10844 0 15603850 7940 0 582934 =
0 0=0A 11659 0 16800328 8503 0 630330 0=
0=0A 11778 0 17033560 8998 0 677934 0 =
0=0A 12149 0 17592134 9504 0 728094 0 0=0A =
12551 0 18223550 9974 0 774164 0 0=0A 1=
3127 0 19093604 10413 0 811858 0 0=0A 13712 =
0 20010140 10924 0 848014 0 0=0A 14499 0=
21153538 11407 0 878252 0 0=0A 14818 0 21=
740270 11979 0 915374 0 0=0A 15831 0 2313644=
6 12376 0 950636 0 0=0A 15912 0 23365454 =
12852 0 997242 0 0=0A 16257 0 23848866 132=
82 0 1041878 0 0=0A 16384 0 24084782 13666 =
0 1079790 0 0=0A 16670 0 24508980 14078 0 =
1106886 0 0=0A 17845 0 26255548 14486 0 113=
4700 0 0=0A 18097 0 26705634 15064 0 1163308 =
0 0=0A 18470 0 27283000 15365 0 1198828 0=
0=0A 18139 0 26842676 15596 0 1225540 0 =
0=0A 18792 0 27799564 16000 0 1264568 0 0=0A =
17854 178 26454106 16521 0 1298714 0 0=0A 1=
6741 1542 24820298 16770 0 1343328 0 0=0A =
input (em0) output=0A packets errs bytes pac=
kets errs bytes colls drops=0A 15288 1667 22683486 17231 =
0 1422690 0 0=0A 15539 1718 23250372 17282 0=
1495058 0 0=0A 14379 545 21541954 17364 0 1=
508696 0 0=0A 14312 1733 21546776 17276 0 150337=
2 0 0=0A 14269 1744 21498908 17516 0 1508294 =
0 0=0A 14444 1729 21766812 17175 0 1482130 0 =
0=0A 15023 1724 22643198 16987 0 1432048 0 0=
=0A 15279 1703 23036294 16909 0 1395094 0 0=0A =
15325 1701 23118536 16938 0 1380268 0 0=0A 15=
572 1684 23494362 16909 0 1344214 0 0=0A 15798 =
1699 23845972 16857 0 1303200 0 0=0A 16195 1683 =
24497790 17059 0 1291586 0 0=0A 16431 1674 248=
51278 16826 0 1245320 0 0=0A 16683 1643 25231910=
16675 0 1204450 0 0=0A 16728 1647 25302534 =
16672 0 1178930 0 0=0A 16826 1649 25455662 1666=
2 0 1178140 0 0=0A 16760 1653 25352830 16480 =
0 1161086 0 0=0A 17002 1634 25720672 16423 0 =
1143508 0 0=0A 16943 1643 25629892 16642 0 1160=
858 0 0=0A 16995 1644 25708823 16539 0 1153782 =
0 0=0A 17026 1643 25758462 16606 0 1153342 0 =
0=0A=0AHowever, network throughput didn't change in the two scenarios a=
bove.=0AIs there anything that I can test to improve my network throughput.=
=0A=0AThanks,=0A=0AWon=0A=0A=0A=0A
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?555333.76711.qm>