netdev - Re: UDP regression with packets rates

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Thu, 10 Sep 2009 23:37:07 +0200
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Christoph Lameter <cl@...ux-foundation.org>
CC:	netdev@...r.kernel.org
Subject: Re: UDP regression with packets rates < 10k per sec

Eric Dumazet a écrit :
> Something must be wrong with program or whatever...
> 
> On the receiver I did this to trace the latency messages only
> 
> # tcpdump -i eth0 not host 239.0.192.2 and port 9002 -n
> 
> 22:28:07.223300 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
> 22:28:07.223403 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32
> 
> 22:28:08.223301 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
> 22:28:08.223416 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32
> 
> 22:28:09.223380 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
> 22:28:09.223494 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32
> 
> 22:28:10.223481 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
> 22:28:10.223597 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32
> 
> 22:28:11.223581 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
> 22:28:11.223678 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32
> 
> 
> See how the answer is *very* slow ? Something like > 100 us ?
> 

Never mind, this is OK, after proper irq affinity and tweaks on receiver :
(I should do same on other machine as well...)

(echo 1 >/proc/irq/default_smp_affinity) before up-ing bnx2 NIC

echo performance >/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

and idle=mwait  (or hpet and C2/C3 states are used on this machine and 
wakeups are slow)

# tcpdump -i eth0 not host 239.0.192.2 and port 9002 -n
23:25:02.137143 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
23:25:02.137212 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32

23:25:03.137204 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
23:25:03.137288 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32

23:25:04.137276 IP 55.225.18.7.9002 > 55.225.18.5.9002: UDP, length 32
23:25:04.137355 IP 55.225.18.5.9002 > 55.225.18.7.9002: UDP, length 32


I presume powersaving improvements in last kernels are against
latencies goals...

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html