On 15.10.2022. 9:39, Stuart Henderson wrote:
> On 2022-10-14, Gabor LENCSE <lencse@hit.bme.hu> wrote:
>> Dear All,
>>
>> I am a researcher and I would like to benchmark the stateful NAT64
>> performance of OpenBSD PF.
>>
>> I use a 32-core server as DUT (Device Under Test). When I use Linux for
>> benchmarking other stateful NAT64 implementations, I use the "ethtool -N
>> enp5s0f1 rx-flow-hash udp4 sdfn" command to include also the source and
>> destination port numbers (not only the source and destination IP
>> addresses) into the hash function to distribute the interrupts caused by
>> packet arrivals evenly among all the CPU cores.
>>
>> I tried to find a similar solution under OpenBSD, but I could not. (I
>> used search expressions like: OpenBSD RSS receive side scaling multi
>> queue receiving) Perhaps it is called differently under OpenBSD, or
>> maybe there is no such solution at all?
>>
>> Could you advise me please?
>
> A few network drivers have support for multiple queues (if my grepping
> is correct: aq igc bnxt ix ixl mcx vmx) - typically you will see the
> nunber of queues reported in the dmesg attach line if supported - but
> there's no interface to adjust what's fed into the hash function.
>
> 32 cores is quite a lot for OpenBSD, more than around 8 is likely to
> be a waste for current versions in many use cases.
>
Hi,
does it make sense to mention RSS and other stuff like TSO, MSI-X,
Multiple queues in man ?
Something like
https://leaf.dragonflybsd.org/cgi/web-man?command=ix§ion=ANY
No comments:
Post a Comment