[Oisf-users] Testers: please test our initial Hyperscan support

[Cooper F. Nelson]

I have this conversation a lot (hyperthreading was invented here at UCSD
so I'm intimately familiar with the technology), but for the vast
majority of modern workloads on modern processors; 'virtual'
hyperthreaded cores are indistinguishable from physical ones.

The only exception is for certain specific instructions/architectures
that share a single execution unit per-core, like floating-point
multiply/divide on the Nephalem micro-architecture.

On our current sensors we are running 16 worker threads on 16 HT cores,
one per core.  We let the kernel schedule the management threads.
Suricata performs much worse with hyper-threading disabled.

-Coop

On 4/6/2016 12:59 AM, Michał Purzyński wrote:
> Our you saying you use as many workers as hyper threads and not
> physical cores (minus some reserved for other Suricata threads). In
> other words - Suricata thread gets a hyper thread?

-- 
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/20160407/d60d4a03/attachment-0002.sig>