FreeBSD 10G forwarding performance @Intel
Alexander V. Chernikov
melifaro at FreeBSD.org
Tue Jul 17 07:32:40 UTC 2012
On 17.07.2012 11:17, Hooman Fazaeli wrote:
> May be slightly off-topic, but do you have tested (or have plans to test )
> with bidirectional traffic?
Situation with bi-directional traffic is better (not sure how much). I'm
intentionally not testing this case to discover rough cases (like
contested interface counters) faster.
Here is statistics from one of the machines running in production
* E5645
* Intel 82599
* HT turned on (so 16 out of 24 cores are used)
* Modified Intel drivers
* 8.3-S kernel with interface rlock patch
* ipfw counters eliminated
* global forwarding counters not eliminated (-another 100-200 kpps)
* route locking eliminated (modified version of original patch in the
first message)
* 8 vlans
* IPv4 traffic
* 4-8 firewall rules to pass
http://static.ipfw.ru/images/degas_pps.png
http://static.ipfw.ru/images/degas_traffic.png
http://static.ipfw.ru/images/degas_cpu.png
--
WBR, Alexander
More information about the freebsd-net
mailing list