aurora-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Maxim Khutornenko" <ma...@apache.org>
Subject Review Request 33487: Bumping up scheduling rate.
Date Thu, 23 Apr 2015 18:01:07 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/33487/
-----------------------------------------------------------

Review request for Aurora and Bill Farner.


Repository: aurora


Description
-------

With the asynchronous preemptor and a series of other perf improvements, our scheduling alrorithm
perfomance improved 230X - 270X:

Before
```
Benchmark                                                                        Mode  Samples
   Score   Error  Units
SchedulingBenchmarks.InsufficientResourcesSchedulingBenchmark.runBenchmark     thrpt     
100  214.098 ± 2.407  ops/s
SchedulingBenchmarks.ConstraintMismatchsSchedulingBenchmark.runBenchmark       thrpt     
100  180.754 ± 2.407  ops/s
```
After
```
Benchmark                                                                      Mode  Cnt 
    Score      Error  Units
SchedulingBenchmarks.InsufficientResourcesSchedulingBenchmark.runBenchmark    thrpt   10 
50328.609 ± 1356.151  ops/s
SchedulingBenchmarks.ValueConstraintMismatchSchedulingBenchmark.runBenchmark  thrpt   10 
49237.475 ± 1803.555  ops/s
```

We are only limited by the storage perf now (native + H2). Real life scheduling round latencies
are hovering around 25msec on average (SSD-backed native log store), which suggests a new
default scheduling rate of 40/sec.


Diffs
-----

  src/main/java/org/apache/aurora/scheduler/async/AsyncModule.java 35c7e43746783ec81254b473848fb8579f9d233e


Diff: https://reviews.apache.org/r/33487/diff/


Testing
-------


Thanks,

Maxim Khutornenko


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