> On Sept. 11, 2018, 8:17 p.m., Meng Zhu wrote:
> > src/tests/hierarchical_allocator_benchmarks.cpp
> > Lines 378 (patched)
> > <https://reviews.apache.org/r/68591/diff/2/?file=2088300#file2088300line378>
> >
> >     As I mentioned in the previous patch, it is confusing here that the 
> > test needs to resume the allocator even it did not pause it. The test will 
> > control the clock, and that should be enough for it to dictate the test 
> > progress.

That was an oversight. Thanks for the catch


> On Sept. 11, 2018, 8:17 p.m., Meng Zhu wrote:
> > src/tests/hierarchical_allocator_benchmarks.cpp
> > Lines 409-411 (patched)
> > <https://reviews.apache.org/r/68591/diff/2/?file=2088300#file2088300line409>
> >
> >     We might as well `unallocate` the offered resources instead of 
> > allocating the `taskResources`, both are not ideal, but with the former, we 
> > at least save some writes:
> >     
> >     ```
> >     Resources remainingResources = offer.resources;
> >     const Resources& taskResources = 
> > frameworkTaskResources.at(offer.frameworkId);
> >     
> >     // We strip allocation information of `remainingResources` so that we
> >     // can compare/subtract with `taskResources`.
> >     remainingResources.unallocate();
> >     
> >     ```

We need to "reallocate" before calling allocator to recover. But yeah, it saves 
some writes.


> On Sept. 11, 2018, 8:17 p.m., Meng Zhu wrote:
> > src/tests/hierarchical_allocator_benchmarks.cpp
> > Lines 436-437 (patched)
> > <https://reviews.apache.org/r/68591/diff/2/?file=2088300#file2088300line436>
> >
> >     in << ... << allocation rounds/cycles

That was a weird mix of wordings. I have updated the patch to address some of 
that.


- Kapil


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/68591/#review208533
-----------------------------------------------------------


On Sept. 13, 2018, 4:25 p.m., Kapil Arya wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/68591/
> -----------------------------------------------------------
> 
> (Updated Sept. 13, 2018, 4:25 p.m.)
> 
> 
> Review request for mesos, Meng Zhu and Till Toenshoff.
> 
> 
> Bugs: MESOS-9187
>     https://issues.apache.org/jira/browse/MESOS-9187
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This tests measures allocation performance with non-uniform framework
> characteristics. Each framework profile launches a different number of
> tasks with different task sizes.
> 
> 
> Diffs
> -----
> 
>   src/tests/hierarchical_allocator_benchmarks.cpp PRE-CREATION 
> 
> 
> Diff: https://reviews.apache.org/r/68591/diff/3/
> 
> 
> Testing
> -------
> 
> ```
> [ RUN      ] HierarchicalAllocations_BENCHMARK_TestBase.Allocations
> Added 80 agents in 17.162452ms
> Added 554 frameworks in 220.842289ms
> Start allocation
> Launched 790 tasks out of 19500 total tasks in round 1 with 80 offers and 
> took 171.384046ms
> Launched 1165 tasks out of 19500 total tasks in round 2 with 79 offers and 
> took 177.912018ms
> Launched 1255 tasks out of 19500 total tasks in round 3 with 78 offers and 
> took 184.146219ms
> Launched 1340 tasks out of 19500 total tasks in round 4 with 77 offers and 
> took 189.107347ms
> Launched 1425 tasks out of 19500 total tasks in round 5 with 76 offers and 
> took 187.070715ms
> Launched 1510 tasks out of 19500 total tasks in round 6 with 75 offers and 
> took 187.07625ms
> Launched 1595 tasks out of 19500 total tasks in round 7 with 74 offers and 
> took 186.33422ms
> Launched 1665 tasks out of 19500 total tasks in round 8 with 73 offers and 
> took 183.589366ms
> Launched 1745 tasks out of 19500 total tasks in round 9 with 72 offers and 
> took 180.402905ms
> Launched 1815 tasks out of 19500 total tasks in round 10 with 71 offers and 
> took 180.515393ms
> Launched 1895 tasks out of 19500 total tasks in round 11 with 70 offers and 
> took 180.966979ms
> Launched 1965 tasks out of 19500 total tasks in round 12 with 69 offers and 
> took 176.934459ms
> Launched 2005 tasks out of 19500 total tasks in round 13 with 68 offers and 
> took 187.481368ms
> Launched 2055 tasks out of 19500 total tasks in round 14 with 67 offers and 
> took 183.099878ms
> Launched 2095 tasks out of 19500 total tasks in round 15 with 66 offers and 
> took 173.505944ms
> Launched 2135 tasks out of 19500 total tasks in round 16 with 65 offers and 
> took 181.777422ms
> Launched 2175 tasks out of 19500 total tasks in round 17 with 64 offers and 
> took 175.975485ms
> Launched 2205 tasks out of 19500 total tasks in round 18 with 63 offers and 
> took 174.766299ms
> Launched 2225 tasks out of 19500 total tasks in round 19 with 62 offers and 
> took 173.047727ms
> Launched 2235 tasks out of 19500 total tasks in round 20 with 61 offers and 
> took 178.637012ms
> Launched 2245 tasks out of 19500 total tasks in round 21 with 60 offers and 
> took 176.6152ms
> Launched 2255 tasks out of 19500 total tasks in round 22 with 59 offers and 
> took 171.762635ms
> Launched 2265 tasks out of 19500 total tasks in round 23 with 58 offers and 
> took 173.781849ms
> Launched 2275 tasks out of 19500 total tasks in round 24 with 57 offers and 
> took 173.308192ms
> Launched 2285 tasks out of 19500 total tasks in round 25 with 56 offers and 
> took 167.873817ms
> Launched 2295 tasks out of 19500 total tasks in round 26 with 55 offers and 
> took 167.150833ms
> Launched 2305 tasks out of 19500 total tasks in round 27 with 54 offers and 
> took 165.603318ms
> Launched 2315 tasks out of 19500 total tasks in round 28 with 53 offers and 
> took 165.629372ms
> Launched 2315 tasks out of 19500 total tasks in round 29 with 52 offers and 
> took 163.471809ms
> ...
> Launched 2315 tasks out of 19500 total tasks in round 530 with 0 offers and 
> took 110.898635ms
> Launched 2315 tasks out of 19500 total tasks in round 531 with 0 offers and 
> took 113.201834ms
> Failed to launch all tasks: Timed out after 1mins
> Resource statistics:
> Cluster capacity: cpus:5120; mem:39040000
> Cluster allocation: cpus:1093.4; mem:1688000
> Target allocation: cpus:2370; mem:17400000
> [       OK ] HierarchicalAllocations_BENCHMARK_TestBase.Allocations (60321 ms)
> ```
> 
> 
> Thanks,
> 
> Kapil Arya
> 
>

Reply via email to