Re: [Intel-wired-lan] [PATCH next-queue v1 0/3] igc: Add support for multiple TX tstamp requests

[Miroslav Lichvar <mlichvar@redhat.com>]

On Thu, Mar 09, 2023 at 02:57:28PM -0800, Vinicius Costa Gomes wrote:
> Vladimir Oltean <vladimir.oltean@nxp.com> writes:
> > If the application never generates multiple requests in flight, then
> > this evaluation is only testing patch 3 (and patches 1 and 2 only to the
> > extent that they don't cause a regression), right?
> >
> 
> That's right. I was more interested in not causing a regression. I could
> run the same test with two (or more) applications and give some numbers,
> but those numbers couldn't be directly compared with the current version
> of the code.
> 
> But good idea. I will change the application to send "batches" of
> packets, so I can configure the number of "in flight" requests.

If you wanted to see the impact on performance and accuracy of
timestamps at the same time, you could use chrony as an NTP server and
generate load with ntpperf. The server can be started as

# chronyd -d local allow 'hwtimestamp eth0' 'clientloglimit 10000000'

ntpperf needs to be started with the -I -H -o options to measure
the difference between server HW TX timestamp and client HW RX
timestamp. The PHC used by ntpperf needs to be synchronized, ideally
via a separate PHC (e.g. by ptp4l+phc2sys).

If the multi-timestamp patches work well, I'd expect the stddev
reported by ntpperf to stay low at a higher rate. See
https://github.com/mlichvar/ntpperf for some examples.

-- 
Miroslav Lichvar

_______________________________________________
Intel-wired-lan mailing list
Intel-wired-lan@osuosl.org
https://lists.osuosl.org/mailman/listinfo/intel-wired-lan