[Oisf-devel] Better PF_RING performance with workers run-mode now instead of autofp

[Chris Wakelin]

I've been running ac/full with workers for weeks (~4000 rules, 6
threads, 40-80kpps on a port-mirrored 1Gb link; I haven't been given my
10Gb mirror back yet), which seems to work OK; the startup time is quite
long, and we do drop some packets (according to PF_RING's own stats in
/proc/net/pf_ring/*-<dev>.*) from time to time.

Best Wishes,
Chris

On 11/01/12 17:49, Martin Holste wrote:
> Update: ac with auto has even worse performance.  Seeing 70-80% loss.
> 
> On Wed, Jan 11, 2012 at 11:30 AM, Martin Holste <mcholste at gmail.com> wrote:
>> Ah, interesting.  I've been using ac-gfbs, though I used to use ac.
>> Since I run "full," the memory footprint is much smaller with gfbs.
>> We have a ton of RAM, so it's the long startup time that's the biggest
>> issue with memory utilization.  I'll flip over to ac with autofp and
>> see if we see any difference.
>>
>> On Wed, Jan 11, 2012 at 1:23 AM, Victor Julien <victor at inliniac.net> wrote:
>>> Are you using ac or ac-gfbs? Anoop has been improving both. I think the
>>> most significant gain would be when using "ac" "single/auto".


-- 
--+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+-
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