aurora-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stephan Erb <s...@apache.org>
Subject Re: Review Request 53829: Make scheduling benchmarks more realistic.
Date Thu, 17 Nov 2016 00:19:56 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/53829/
-----------------------------------------------------------

(Updated Nov. 17, 2016, 1:19 a.m.)


Review request for Aurora, Mehrdad Nurolahzade and Zameer Manji.


Bugs: AURORA-1802
    https://issues.apache.org/jira/browse/AURORA-1802


Repository: aurora


Description (updated)
-------

This patch aims to strike a better balance between available offers, already
launched tasks, and tasks to be scheduled. Specifically, it ensures that we
only measure scheduling of tasks for which instances of the same job are
already running (e.g. as needed by limit constraints).

The following setup is now the default: Given N hosts

* there are 0.25 * N tasks of job A scheduled
* there are 0.25 * N tasks of job B scheduled
* there are 0.50 * N free offers available
* we try to schedule a task instance of job A

Hopefully this provides us with benchmark results closer to actual
production performance.

Before:

SchedulingBenchmarks.LimitConstraintMismatchSchedulingBenchmark.runBenchmark
thrpt   10  1646.245 ± 273.340  ops/s

After:

SchedulingBenchmarks.LimitConstraintMismatchSchedulingBenchmark.runBenchmark
thrpt   10  241.808 ± 51.952  ops/s


Diffs
-----

  src/jmh/java/org/apache/aurora/benchmark/BenchmarkSettings.java 94f8b799c15601758aed1c5458ad7af79bda6bbb

  src/jmh/java/org/apache/aurora/benchmark/SchedulingBenchmarks.java 1b56500260fc6539facf0ae3a25db02cc72b82db


Diff: https://reviews.apache.org/r/53829/diff/


Testing
-------


Thanks,

Stephan Erb


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