[Oisf-devel] 77b708=WIN?
Chris Wakelin
c.d.wakelin at reading.ac.uk
Tue Sep 27 15:29:13 UTC 2011
I like it because it's simple and I've tied each one of my 6 threads to
a particular CPU (set_cpu_affinity: yes).
Will comments in Issue 315
(https://redmine.openinfosecfoundation.org/issues/315) that performance
will be worse than autofp with one or two top talkers (which we have at
times - generally SSH between some of our researchers and supercomputing
clusters - a prime candidate perhaps for BPF filtering :) ). However,
load does seem pretty well balanced most of the time, e.g.:
> Cpu0 : 2.6%us, 3.0%sy, 0.0%ni, 92.8%id, 0.0%wa, 0.0%hi, 1.6%si, 0.0%st
> Cpu1 : 13.8%us, 1.9%sy, 0.0%ni, 82.6%id, 0.0%wa, 0.0%hi, 1.6%si, 0.0%st
> Cpu2 : 7.6%us, 2.3%sy, 0.0%ni, 89.1%id, 0.0%wa, 0.0%hi, 1.0%si, 0.0%st
> Cpu3 : 16.1%us, 0.3%sy, 0.0%ni, 82.6%id, 0.0%wa, 0.0%hi, 1.0%si, 0.0%st
> Cpu4 : 10.4%us, 2.3%sy, 0.0%ni, 85.6%id, 0.0%wa, 0.0%hi, 1.7%si, 0.0%st
> Cpu5 : 9.2%us, 1.0%sy, 0.0%ni, 88.7%id, 0.0%wa, 0.0%hi, 1.0%si, 0.0%st
> Cpu6 : 11.1%us, 0.7%sy, 0.0%ni, 87.3%id, 0.0%wa, 0.0%hi, 1.0%si, 0.0%st
> Cpu7 : 2.5%us, 1.9%sy, 0.0%ni, 94.9%id, 0.0%wa, 0.0%hi, 0.6%si, 0.0%st
> Mem: 16465276k total, 15615844k used, 849432k free, 23568k buffers
> Swap: 3906552k total, 57612k used, 3848940k free, 8889352k cached
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ P COMMAND
> 19502 snort 20 0 360m 280m 3028 R 20.2 1.7 30:11.24 2 argus
> 16754 root 20 0 5791m 5.4g 49m S 16.6 34.1 29:41.79 3 Pfring3
> 16752 root 20 0 5791m 5.4g 49m S 13.3 34.1 30:05.45 1 Pfring1
> 16755 root 20 0 5791m 5.4g 49m S 11.3 34.1 24:06.49 4 Pfring4
> 16757 root 20 0 5791m 5.4g 49m R 9.9 34.1 25:30.74 6 Pfring6
> 16756 root 20 0 5791m 5.4g 49m S 7.6 34.1 27:50.01 5 Pfring5
> 16753 root 20 0 5791m 5.4g 49m S 6.3 34.1 23:36.11 2 Pfring2
> 16758 root 20 0 5791m 5.4g 49m S 3.3 34.1 11:16.54 7 FlowManagerThre
> 19501 snort 20 0 360m 280m 3028 S 1.0 1.7 1:06.31 2 argus
(the "snort" uid is a - ahem - historical artifact :) )
There's a possibility I made a mess of including Anoop's changes though.
I've attached my version of the patch against git master. Does it look
OK? Having said that, I'd expect it to crash a lot more if it's wrong.
Best Wishes,
Chris
On 27/09/11 16:14, Will Metcalf wrote:
> Yes, that is correct, Victor essentially told me that naming it
> "single" was dumb and confusing but it in a really polite accent name
> /me looks over at large dutch guy in his living room.. I'm operating
> under the assumption that victor is practicing the art of middle
> kingdom diplomacy :)..
>
> Regards,
>
> Will
>
> On Tue, Sep 27, 2011 at 10:11 AM, Martin Holste <mcholste at gmail.com> wrote:
>>> Maybe I just misunderstand you, but runmode single is using a single
>>> packet processing pipeline. If you need more I suggest autofp.
>>>
>> If I understood Will correctly, the single runmode for PF_RING was
>> because PF_RING already does all of the flow distribution, so you want
>> as close to n number of singles as possible to keep all of the threads
>> related to a given flow on the proper CPU. I don't know if that's
>> been corrected in autofp now.
>> _______________________________________________
>> Oisf-devel mailing list
>> Oisf-devel at openinfosecfoundation.org
>> http://lists.openinfosecfoundation.org/mailman/listinfo/oisf-devel
>>
> _______________________________________________
> Oisf-devel mailing list
> Oisf-devel at openinfosecfoundation.org
> http://lists.openinfosecfoundation.org/mailman/listinfo/oisf-devel
--
--+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+-
Christopher Wakelin, c.d.wakelin at reading.ac.uk
IT Services Centre, The University of Reading, Tel: +44 (0)118 378 2908
Whiteknights, Reading, RG6 6AF, UK Fax: +44 (0)118 975 3094
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pf_ring_single.patch
Type: text/x-patch
Size: 4557 bytes
Desc: not available
URL: <http://lists.openinfosecfoundation.org/pipermail/oisf-devel/attachments/20110927/485295dd/attachment.bin>
More information about the Oisf-devel
mailing list