[Framework-Team] Re: PLIP load test reports available

Ross Patterson me at rpatterson.net
Sun Sep 6 16:36:19 UTC 2009


Ross Patterson <me at rpatterson.net> writes:

> Maurits van Rees <maurits at vanrees.org>
> writes:
>
>> On Sun, Sep 06, 2009 at 05:52:52PM +0800, Martin Aspeli wrote:
>>> This is really good stuff!
>>>
>>> I must admit, I don't understand the graphs or the tables-of-graphs
>>> at all, though. Is there a quick explanation somewhere about how to
>>> read them?
>>
>> The reference bench results are for Plone4 without plips.
>> The challenger bench results are for one of the plips.
>>
>> If the graph is mostly green, the plip has better performance.
>> If the graph is mostly red, the plip has worse performance.
>
> Thanks for the documentation, very helpful!
>
>> For each plip three tests have been done, showing performance for:
>> - content creation
>> - read only
>> - heavy writes
>> I don't know what is being tested exactly.
>
> See the tests package in collective.coreloadtests.  Should be very
> readable:
>
> http://dev.plone.org/collective/browser/collective.coreloadtests/trunk/collective/coreloadtests/tests
>
> Also, using the funkload test recorder makes creating new tests pretty
> easy:
>
> http://funkload.nuxeo.org/#test-recorder
>
>> On the front page of the tests at
>> http://weblion.psu.edu/static/loadtesting/plone4.0/plips.htmlyou have per plip the following items:
>> - label/name of the plip
>> - for each of the three tests:
>>   - main graph, with link to detailed report of the plip
>>   - main difference graph between plip and plain plone, with link to
>>     detailed difference report
>>
>> Take the diffence page for plip 9310 showing heavy writes:
>> http://bit.ly/vN9kJ>
>> The first graph has the requests per second.
>> The blue line shows the results for B1 (bench 1, plain plain 4.0)
>> The purple line shows the results for B2 (bench 2, plip 9310)
>>
>> The first part shows the difference between those lines as red,
>> meaning that the plip is slightly slower; this part of the graph is
>> for 1-3 concurrent users.
>>
>> To the right we have a green difference, meaning that the plip is
>> faster, even up to 80% for 10 concurrent users.
>
> Actually, this is one of the invalid tests.  If you look at the test
> data you see that it has 100% test failure.  This is probably why the
> through put was so high.
>
> http://weblion.psu.edu/static/loadtesting/plone4.0/test_WriteHeavy-20090906T002920-plip9310-flexible-user-registration/index.html#test-stats
>
> So the things that should raise a flag are differential graphs with
> solid color (red or green) between the curve and the X axis, and
> individual test report graphs with a horizontal red line at the top of
> the bottom portion (the error portion, only present if there are hours),
> of the per-test report image.

Another with problems:

http://weblion.psu.edu/static/loadtesting/plone4.0/test_WriteHeavy-20090906T044758-plip9330-choose-member-role/index.html#test-stats

Ross

>> If results look too absurd, probably something went wrong in the
>> tests.  For example plip 8814, replacing secure mail host, looks
>> totally red.  Apparently with that plip we serve a whopping zero pages
>> per second. :-)
>
> Exactly, and that's where the *.log files come in.  Also, clicking
> through to the individual test report might also be informative.
>
>> I hope that clears things up a bit.
>
> Greatly, I think.  Thanks so much.
>
> Ross





More information about the Framework-Team mailing list