IPI Benchmark Data?

Sepherosa Ziehau sepherosa at gmail.com
Tue Aug 2 19:35:55 PDT 2016


On Wed, Aug 3, 2016 at 10:27 AM, Alex Merritt <merritt.alex at gmail.com> wrote:
> Hello all,
>
> I saw 4.6 was released, great! From the release notes:
>
>> SMP performance was already very good. As part of the NVMe driver work we
>> revamped the buffer cache subsystem and a number of other I/O related paths,
>> further reducing lock contention and IPI signalling overheads.
>
> Was the IPI signaling cost reduced due to the cache improvements, or were
> there other updates specifically for the IPI subsystem to improve the
> performance?
>
> Is there any available data illustrating or describing the IPI latencies,
> overheads, sustained throughput, or other characteristics on DF across cores
> and sockets on SMP systems?

There are set of sysctls to test IPI latency:

debug.ipiq.latency_test
debug.ipiq.latencyX

test/sysperf/ipitest uses these sysctls to test IPI latency in various
MWAIT C-states.

Thanks,
sephe

-- 
Tomorrow Will Never Die



More information about the Kernel mailing list