[Swan-dev] test run 80 min on 2 x Xeon E5-2620L(7 years old server)

Antony Antony antony at phenome.org
Thu Apr 11 15:46:54 UTC 2019


On Thu, Apr 11, 2019 at 10:21:52AM -0400, Andrew Cagney wrote:
> On Thu, 11 Apr 2019 at 03:17, Antony Antony <antony at phenome.org> wrote:
> >
> > Toumo brought up his old server, running F29, and I am doing a benchmark of
> > testrun - time run our current KVM test suite using kvmrunner.
> >
> > Here are the first results.
> > 80 Minutes for a testrun. This is impressive to me.
> > Number of tests from master ~748.
> >

I ran about 7-10 runs one run stopped at 386. I didn't capture the logs.
the rest seems stable. 

I tried another combination and 24/40 runtime 55 minutes for 748 tests. 

|KVM_WORKERS/KVM_PREFIX| run time |
| 12/28 | 79 Minutes |
| 8/24 |  101 |
| 8/28 | 118 |
|24/40 | 55 | 

I think the CPU is good at switching tasks with less issues.

> Are the actual results relatively stable?

I guss so. What signs for unreliable runs?
I noticed "timeout": 12 I see the same on testing.libreswan.org too.

> I only tested a machine with cores==threads but you've got
> cores<threads so you might want to add the results to

the current testing.libreswan.org is also cores<threads? 4<8
This one 16<32 and two CPU dies.


> https://libreswan.org/wiki/Test_Suite_-_Performance
> 
> > I suspect there is still room tune more.  We will see.
> >
> > So the 7 year old Enterprise Xeon perform better than a new (6 months old)
> > basic servers in the same price range.
> 
> To be fair, I've not tried to tune the new server, I've been waiting
> for the failures to settle down (#include sound of drumming fingers)
> (somewhere there's a todo to embed the above numbers in the web
> results)

I am curious how far we can push the current testing.libreswan.org
current runs of ~5+ hours looks poor to me.

While this old server is able to do a run less than an hour.

-antony



More information about the Swan-dev mailing list