Re: [PATCH v2 0/4] locking/mutex: Enable optimistic spinning of lock waiter

From: Waiman Long
Date: Mon Mar 21 2016 - 23:19:17 EST

On 02/16/2016 03:51 AM, Peter Zijlstra wrote:
On Fri, Feb 12, 2016 at 12:32:11PM -0500, Waiman Long wrote:
My own test on a 4-socket E7-4820 v3 system showed a regression of
about 4% in the high_systime workload with Peter's patch which this
new patch effectively eliminates.

Testing on an 8-socket Westmere-EX server, however, has performance
change from -9% to than +140% on the fserver workload of AIM7
depending on how the system was set up.
Subject: [lkp] [locking/mutex] aaca135480: -72.9% fsmark.files_per_sec

My patch also generated the above email.

Please also test that benchmark against this approach.

I also got an email from "kernel test robot", it didn't list fsmark at all. Instead, the subject was

[lkp] [locking/mutex] 5267438002: +38.9% fileio.time.involuntary_context_switches

4409 ± 1% +38.9% 6126 ± 2% fileio.time.involuntary_context_switches
6.00 ± 0% +33.3% 8.00 ± 0% fileio.time.percent_of_cpu_this_job_got
36.06 ± 0% +43.0% 51.55 ± 0% fileio.time.system_time
1828660 ± 0% -92.5% 137258 ± 0% fileio.time.voluntary_context_switches

Given that the number of voluntary context switches dropped by 92.5%, an increase in involuntary context switches that is order of magnitude less than the voluntary context switches should be OK, I think.

Do you know how to report back that this increase is expected and is nothing to worry about? Do I just reply it back?