[
https://issues.apache.org/jira/browse/FLINK-15171?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16992587#comment-16992587
]
Roman Grebennikov commented on FLINK-15171:
-------------------------------------------
[~pnowojski] I've tried running these benchmarks on my hardware and got weird
results:
{noformat}
1thread master SerializationFrameworkMiniBenchmarks.serializerTuple thrpt 100
608.128 ± 8.701 ops/ms
1thread no-pr SerializationFrameworkMiniBenchmarks.serializerTuple thrpt 100
605.246 ± 9.584 ops/ms
4thread no-pr SerializationFrameworkMiniBenchmarks.serializerTuple thrpt 100
610.825 ± 10.945 ops/ms
4thread master SerializationFrameworkMiniBenchmarks.serializerTuple thrpt 100
613.504 ± 8.557 ops/ms{noformat}
TLDR: no changes, as in original PR.
I may suspect that the difference may come out of different memory access
patterns in the old and new version of serializers on different hardware: for
example, my desktop having better memory throughput compared to the one used on
dak8s.net.
Can you please describe a bit more the hardware on this Hetzner box used for
benchmarking:
* CPU model (cat /proc/cpuinfo)
* memory details (cat /proc/meminfo)
* hardware details (sudo dmidecode)
So I can try to reproduce the benchmark problem.
> Performance regression in serialisation benchmarks
> --------------------------------------------------
>
> Key: FLINK-15171
> URL: https://issues.apache.org/jira/browse/FLINK-15171
> Project: Flink
> Issue Type: Bug
> Components: API / Type Serialization System, Benchmarks
> Affects Versions: 1.10.0
> Reporter: Piotr Nowojski
> Priority: Blocker
> Fix For: 1.10.0
>
>
> There is quite significant performance regression in serialisation benchmarks
> in the commit range 2ecf7ca..9320f34 (which includes FLINK-14346).
> http://codespeed.dak8s.net:8000/timeline/?ben=serializerTuple&env=2
> http://codespeed.dak8s.net:8000/timeline/?ben=serializerRow&env=2
> http://codespeed.dak8s.net:8000/timeline/?ben=serializerPojo&env=2
> it coincides with the performance improvement for heavy strings
> http://codespeed.dak8s.net:8000/timeline/?ben=serializerHeavyString&env=2
> it might be caused by some accidental change in the benchmarking code
> (changing parallelism in one benchmarks is carried on to the next one?) or in
> the code itself.
> CC [~rgrebennikov] [~AHeise]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
