aurora-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bill Farner" <wfar...@apache.org>
Subject Re: Review Request 31508: Removing redundant scheduling loop in preemptor.
Date Fri, 27 Feb 2015 17:32:20 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/31508/#review74514
-----------------------------------------------------------

Ship it!


Context is that the removed code was attempting to mitigate a race where we preempt a task
unnecessarily because new offers have arrived.  Given that preemption begins immediately after
we have passed over the offers, this additional pass is redundant.  Additionally, the race
still exists (an offer could arrive _right after_ we preempt), so the mitigation is of dubious
value.

- Bill Farner


On Feb. 27, 2015, 12:04 a.m., Maxim Khutornenko wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/31508/
> -----------------------------------------------------------
> 
> (Updated Feb. 27, 2015, 12:04 a.m.)
> 
> 
> Review request for Aurora, Bill Farner and Zameer Manji.
> 
> 
> Bugs: AURORA-1156
>     https://issues.apache.org/jira/browse/AURORA-1156
> 
> 
> Repository: aurora
> 
> 
> Description
> -------
> 
> This is #1 from the attached ticket. Brings anywhere between 2% and 18% better perf in
bechmark scenarios.
> 
> BEFORE:
> ```
> Benchmark                                                                           
   Mode  Samples         Score         Error  Units
> o.a.a.b.SchedulingBenchmarks.InsufficientResourcesSchedulingBenchmark.runBenchmark  
   avgt      100    909677.646 ±   10103.466  ns/op
> o.a.a.b.SchedulingBenchmarks.LimitConstraintMismatchSchedulingBenchmark.runBenchmark
   avgt      100   1332768.205 ±   16664.386  ns/op
> o.a.a.b.SchedulingBenchmarks.PreemptorFallbackForLargeClusterBenchmark.runBenchmark 
   avgt      100  69304405.590 ± 1536571.317  ns/op
> o.a.a.b.SchedulingBenchmarks.ValueConstraintMismatchSchedulingBenchmark.runBenchmark
   avgt      100    870348.707 ±   16815.495  ns/op
> ```
> 
> AFTER:
> ```
> Benchmark                                                                           
   Mode  Samples         Score         Error  Units
> o.a.a.b.SchedulingBenchmarks.InsufficientResourcesSchedulingBenchmark.runBenchmark  
   avgt      100    749864.522 ±    6568.372  ns/op
> o.a.a.b.SchedulingBenchmarks.LimitConstraintMismatchSchedulingBenchmark.runBenchmark
   avgt      100   1125995.085 ±   19241.796  ns/op
> o.a.a.b.SchedulingBenchmarks.PreemptorFallbackForLargeClusterBenchmark.runBenchmark 
   avgt      100  68028627.539 ± 1412569.919  ns/op
> o.a.a.b.SchedulingBenchmarks.ValueConstraintMismatchSchedulingBenchmark.runBenchmark
   avgt      100    766747.584 ±   13310.142  ns/op
> ```
> 
> 
> Diffs
> -----
> 
>   src/main/java/org/apache/aurora/scheduler/async/preemptor/PreemptorImpl.java a4e8dd3b9b58dac2d8507042500b7a479d46ebc0

>   src/test/java/org/apache/aurora/scheduler/async/preemptor/PreemptorImplTest.java 44cd8f79493f0f247cd876ef78b30b4f813314c4

> 
> Diff: https://reviews.apache.org/r/31508/diff/
> 
> 
> Testing
> -------
> 
> ./gradlew jmh
> 
> 
> Thanks,
> 
> Maxim Khutornenko
> 
>


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