[Oisf-users] af-packet and Linux Kernel version

Cooper F. Nelson cnelson at ucsd.edu
Tue Nov 15 20:23:30 UTC 2016

Hi Michal (and great presentation!),

Turned off and verified:

> receive-hashing: off

I tried running with a single RSS queue pinned to core 0 and suricata
running worker threads on cores 1-15, however core 0 is still at 100%
utilization and there was massive packet loss on the NIC/bus (this will
not show in stats.log, btw).

I tried again using cores 0-1 and suri on 2-14.  This working better and
we are seeing lots of alerts, so maybe this is the best we can do on
current hardware.  We will still have packet re-ordering due to timing
issues and associated missed events, but it will be less than the other
modes I think.

If anyone has any tips on how to max out single-core RSS performance,
please let me know.  We are seeing 800k packets/sec peak, which is well
under what Michal and Peters were seeing in their experiments, so I'm
not sure what the issue is.  It's an older system, but still fairly beefy:

> processor       : 0
> vendor_id       : GenuineIntel
> cpu family      : 6
> model           : 26
> model name      : Intel(R) Xeon(R) CPU           X5560  @ 2.80GHz
> stepping        : 5
> microcode       : 0x11
> cpu MHz         : 2792.918
> cache size      : 8192 KB
> bogomips        : 5585.83


On 11/15/2016 11:01 AM, Michał Purzyński wrote:
> The new afpacket from 4.4 will use card hash if rxhash variable is enabled.
> Disable it with ethtool and verify with ethtool -k 

Cooper Nelson
Network Security Analyst
UCSD ITS Security Team
cnelson at ucsd.edu x41042

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
URL: <http://lists.openinfosecfoundation.org/pipermail/oisf-users/attachments/20161115/5579be27/attachment-0002.sig>

More information about the Oisf-users mailing list