> On Feb. 27, 2015, 9:32 a.m., Bill Farner wrote:
> > 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.

Maxim, can you put this additional context in the commit message?


- Zameer


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


On Feb. 26, 2015, 4:04 p.m., Maxim Khutornenko wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/31508/
> -----------------------------------------------------------
> 
> (Updated Feb. 26, 2015, 4:04 p.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
> 
>

Reply via email to