Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 12 Aug 2009 13:29:35 +0100 (BST)
From:      Robert Watson <rwatson@FreeBSD.org>
To:        current@FreeBSD.org
Subject:   Call for regression and performance testing - 8.0
Message-ID:  <alpine.BSF.2.00.0908121329130.66335@fledge.watson.org>

next in thread | raw e-mail | index | archive | help

Dear all:

As we're approaching 8.0 BETA3, now would be a really good time to start 
identifying functional and performance regressions from the 7.x series. We've 
done a mixed job at this in the past, but when it comes to "I'll do better some 
day", there's no time like the present :-).  Some notes:

   Functional testing

We actually have a sizable regression test suite in src/tools/regression -- 
some of these tools will have broken as a result of 8-CURRENT development, so 
the first task may be to fix them.  The next is to run them on 7-STABLE and 
8-CURRENT and decide if things have gotten worse -- or maybe we have a bug to 
fix in both.  Pick the tool of your choice, and give it a spin.  More than one 
person per tool is fine, because that way we get more diverse testing.

   Performance testing

On the performance front, life is always a bit more tricky -- performance 
testing is a subtle art.  However, the most clear lessons are that (a) testing 
with diverse workloads and diverse environments is extremely important, (b) you 
should do multiple runs and use ministat(1) to analyze results, and (c) that 
you want to compare apples with apples -- use the same hardware/configuration 
wherever possible.  Watch out for annoying nits such as partition layout 
affecting I/O throughput for two different installs on the same disk.

Pick something you think is important to you: bytes/sec over TCP on loopback, 
web hits/sec, NFS ops/sec, disk I/O transactions/sec, and do some comparison 
between 7.2 and 8.0.  If you find improvement -- great!  If you find a 
regression, please start a thread on current@ to help get it diagnosed.  And if 
you want help doing performance measurement for a particular workload that 
isn't well studied, send some e-mail to performance@ to ask for advice on how 
best to measure it.

Thanks,

Robert N M Watson
Computer Laboratory
University of Cambridge



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?alpine.BSF.2.00.0908121329130.66335>