[
https://issues.apache.org/jira/browse/MESOS-1018?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13940063#comment-13940063
]
Benjamin Mahler commented on MESOS-1018:
----------------------------------------
The benchmark framework listed above may be helpful for micro-benchmarking some
of the code in libprocess, it would be beneficial to assess how well it deals
with asynchronous code and multiple threads.
But the primary motivation here is to see macro-benchmarks to assess the
performance of our Actor model. You're asking the right questions with respect
to Scala and Erlang, would be great to do some research for how they benchmark
and analyze performance.
Scalability: very interesting to us, how many Actors can the system support?
Are there performance bottlenecks for a large number of Actors? Can we improve
the scheduling to multiplex Actors onto threads? (For example, the Mesos Master
currently uses an Actor for each slave that it health checks, this could be
tens of thousands of Actors).
Memory Usage: Less interesting, but still great to know what the memory growth
looks like with respect to # Actors.
As to which framework to use for the macro-benchmarking, we could use the one
listed above if it works well for this purpose, otherwise we can just re-use
our existing google-test tooling to produce a separate benchmark test binary.
> Implement Libprocess Benchmark Suite
> ------------------------------------
>
> Key: MESOS-1018
> URL: https://issues.apache.org/jira/browse/MESOS-1018
> Project: Mesos
> Issue Type: Improvement
> Reporter: Dave Lester
> Labels: gsoc2014
>
> Implement a benchmark suite for libprocess to identify potential performance
> improvements and test for performance regressions.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
