mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Anindya Sinha <anindya_si...@apple.com>
Subject Re: Review Request 49571: Added a benchmark test for allocations.
Date Wed, 20 Jul 2016 00:04:46 GMT


> On July 19, 2016, 9:45 p.m., Jiang Yan Xu wrote:
> > Commented on the resources benchmarks. Also let's pull it out then we can hopefully
commit the arithmetic operations for shared resources patch first.
> > 
> > On the separate review could you post the numbers from cout instead of the total
test time?
> > 
> > Overall I think this benchmark is sufficient for a first cut while we are thinking
about more sophisticated tests that evaluate the performance of shared persistent volumes
fairly.
> > 
> > In the new review perhaps add klaus1982 as the review as they are writing benchmarks
for regular DiskInfo/Persistent Volumes (I pinged @klaus1982 on #allocator channel on slack)
> > 
> > If they haven't done so, we can add a benchmark for regular persistent volumes:
add up (e.g.,) 5000 distinct persistent volumes together and see the performance. But then,
what we can achieve is to compare the performance of regular persistent volumes with and without
the new patch: to verify that the new patch doesn't lead to degradation of regular persistent
volume arithmetic.

Moves resources benchmark to https://reviews.apache.org/r/50205/.


> On July 19, 2016, 9:45 p.m., Jiang Yan Xu wrote:
> > src/tests/resources_tests.cpp, lines 2592-2603
> > <https://reviews.apache.org/r/49571/diff/6/?file=1445905#file1445905line2592>
> >
> >     Can we use `createPersistentVolume()` to create the `disk` directly?

Creating the shared persistent volume via `createDiskResource()`.


- Anindya


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/49571/#review142771
-----------------------------------------------------------


On July 19, 2016, 10:53 p.m., Anindya Sinha wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/49571/
> -----------------------------------------------------------
> 
> (Updated July 19, 2016, 10:53 p.m.)
> 
> 
> Review request for mesos and Jiang Yan Xu.
> 
> 
> Bugs: MESOS-5771
>     https://issues.apache.org/jira/browse/MESOS-5771
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Allocations test has the following configurations:
> (1) REGULAR: Offers from every slave have regular resources.
> (2) SHARED: Offers from every slave include a shared resource.
> (3) REGULAR: Offers from every alternate slave contain only regular
>     resources; and offers from every other alternate slave contains
>     a shared resource.
> 
> 
> Diffs
> -----
> 
>   src/tests/hierarchical_allocator_tests.cpp 3ddce7ab19613831527f010524b8454fecfb9927

> 
> Diff: https://reviews.apache.org/r/49571/diff/
> 
> 
> Testing
> -------
> 
> All tests passed.
> 
> Allocations benchmark test results
> ==================================
> Support of shared resources has a small impact on runtime performance in allocations.
Also, there is no visible impact in performance when shared resources are added in the tests.
> 
> With the patch (and no shared resources)
> ----------------------------------------
> round 0 allocate took 3.19704secs to make 200 offers
> round 50 allocate took 3.240605secs to make 200 offers
> round 100 allocate took 3.227024secs to make 200 offers
> round 150 allocate took 3.225281secs to make 200 offers
> round 199 allocate took 3.26036secs to make 200 offers
> 
> With the patch (and shared resources on all agents)
> ---------------------------------------------------
> round 0 allocate took 3.279115secs to make 200 offers
> round 50 allocate took 3.273396secs to make 200 offers
> round 100 allocate took 3.278509secs to make 200 offers
> round 150 allocate took 3.275959secs to make 200 offers
> round 199 allocate took 3.278151secs to make 200 offers
> 
> With the patch (and shared resources on alternate agents)
> ---------------------------------------------------------
> round 0 allocate took 3.251739secs to make 200 offers
> round 50 allocate took 3.263777secs to make 200 offers
> round 100 allocate took 3.263079secs to make 200 offers
> round 150 allocate took 3.263114secs to make 200 offers
> round 199 allocate took 3.236228secs to make 200 offers
> 
> Based on HEAD, with all regular resources (no shared resources in HEAD supported)
> ---------------------------------------------------------------------------------
> round 0 allocate took 2.925681secs to make 200 offers
> round 50 allocate took 2.922036secs to make 200 offers
> round 100 allocate took 2.909337secs to make 200 offers
> round 150 allocate took 2.914093secs to make 200 offers
> round 199 allocate took 2.923762secs to make 200 offers
> 
> 
> Thanks,
> 
> Anindya Sinha
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message