Re: [Audit-test-developer] suite speedups, test fixes, generic suite improvements

[Linda K. <lin...@hp...>]

Wow, this is really impressive.  Thanks for sorting out the quirks
and bugs in the test suite that the sleeps mostly worked around.
I recall some of these sleeps being added by the IBMers as they were
testing with ppc and s390.  Have the tests been run on those
architectures?

In any case, I think Miroslav should go ahead and push these patches.
This is a huge improvement.

Thanks again,

-- ljk

On 10/07/13 07:27, Jiri Jaburek wrote:
> Hi,
> yet another batch of changes from our team is here.
> 
> This time, it's mostly about making the suite faster, with related
> changes all around the idea. Aside from those, a new "make rerun"
> feature is included, which re-runs only non-PASSed tests, along with
> a few generic fix-ups.
> 
> There are several important things I'd like to point out.
> First, the discussed issue of "making lblnet_tst_server inetd-only"
> no longer exists, I managed to create a solution which retains the
> original standalone daemon functionality, second, the
> "TCP RST related tests" change is included, and third, there's one
> more similar change in this patchset I would like to point out
> explicitly - patch 15. I'd really like some comments on that one.
> 
> All changes are RHEL-6.2 compatible, I've tested both base and mls
> runs of the suite without fails or errors.
> 
> 
> Since this patch series is mainly about suite speedups, I should
> probably provide some benchmarks. When trying to generate those,
> I encountered an issue with tcp_syn_retries behaving differently
> on RHEL6.2, RHEL6.3+ and RHEL7 (described in patch 04), so I had
> to do three separate runs.
> The following is a "time make run" of the three networking-related
> buckets combined (network, netfilter, netfilebt):
> 
> RHEL6.2,  default tcp_syn_retries, upstream suite = 101 minutes
> RHEL6.2,  default tcp_syn_retries, patched suite  = 43  minutes
> RHEL6.2,  custom  tcp_syn_retries, patched suite  = 37  minutes
> 
> RHEL6.3+, default tcp_syn_retries, upstream suite = 118  minutes
> RHEL6.3+, default tcp_syn_retries, patched suite  = 69  minutes
> RHEL6.3+, custom  tcp_syn_retries, patched suite  = 37  minutes
> 
> and based on known default value of tcp_syn_retries on RHEL7, we can
> simulate a RHEL7 full-pass run on RHEL6.3+:
> 
> (RHEL7),  default tcp_syn_retries, upstream suite = 136 minutes
> (RHEL7),  default tcp_syn_retries, patched suite  = 90  minutes
> (RHEL7),  custom  tcp_syn_retries, patched suite  = 37  minutes
> 
> IOW, this patchset speeds up things on RHEL6.2 by 64 minutes,
> on RHEL6.3+ by 81 minutes and on RHEL7 by 99 minutes (or 1h39m),
> counting only the networking-related buckets, on our hardware.
> More speed improvements come from patches 07 and 22.
> 
> 
> Please see commit messages of respective patches for more information,
> the patches are attached via In-Reply-To/References to this mail.
> 
> Thanks for the review,
> Jiri
> 
> ------------------------------------------------------------------------------
> October Webinars: Code for Performance
> Free Intel webinars can help you accelerate application performance.
> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from 
> the latest Intel processors and coprocessors. See abstracts and register >
> http://pubads.g.doubleclick.net/gampad/clk?id=60134791&iu=/4140/ostg.clktrk
> _______________________________________________
> Audit-test-developer mailing list
> Aud...@li...
> https://lists.sourceforge.net/lists/listinfo/audit-test-developer
>