Re: [igt-dev] ✗ Fi.CI.IGT: failure for tests/perf_pmu: Avoid RT thread for accuracy test (rev3)

From: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
To: igt-dev@lists.freedesktop.org,
	Patchwork <patchwork@emeril.freedesktop.org>,
	Tvrtko Ursulin <tursulin@ursulin.net>
Subject: Re: [igt-dev] ✗ Fi.CI.IGT: failure for tests/perf_pmu: Avoid RT thread for accuracy test (rev3)
Date: Wed, 4 Apr 2018 12:13:12 +0100	[thread overview]
Message-ID: <40ae9217-8063-f7f0-de79-9c3f8c811248@linux.intel.com> (raw)
In-Reply-To: <20180403183306.2175.85776@emeril.freedesktop.org>

On 03/04/2018 19:33, Patchwork wrote:
> == Series Details ==
> 
> Series: tests/perf_pmu: Avoid RT thread for accuracy test (rev3)
> URL   : https://patchwork.freedesktop.org/series/40662/
> State : failure

[snip]

> shard-apl        total:3498 pass:1835 dwarn:1   dfail:0   fail:7   skip:1655 time:12922s
> shard-hsw        total:3498 pass:1782 dwarn:1   dfail:0   fail:4   skip:1710 time:11552s
> shard-snb        total:3498 pass:1372 dwarn:1   dfail:0   fail:16  skip:2109 time:7043s
> Blacklisted hosts:
> shard-kbl        total:3498 pass:1960 dwarn:1   dfail:0   fail:7   skip:1530 time:9398s

It all passed with quite stable results here.

One thing which is now visible with added variance reporting (even 
though I forgot to divide it by N), is that KBL suffers from much higher 
variance compared to APL.

APL manages to calibrate the loop to 0.01 - 0.08% variance, while KBL 
only manages 0.15 - 0.53%. (Need to divide by N, but relative comparison 
is fine.)

I don't know what to think since normally I'd expect APL to be more 
jittery. Clock source related messages during boot look similar on both. 
Even though they both complain about unstable clock.

Anyways, this much larger variance is perhaps a clue to explain sporadic 
failures on KBL.

Regards,

Tvrtko
_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev