[
https://issues.apache.org/jira/browse/IGNITE-6123?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16275860#comment-16275860
]
Oleg Ignatenko commented on IGNITE-6123:
----------------------------------------
Hi [~avinogradov],
I think I addressed all suggestions you mentioned so far. Specifically:
I changed the way how data was initialised so that each execution now runs
modified data (I figured how to do changes in a lightweight way so that it
doesn't have a substantial impact on performance measurements).
I reviewed config values and found the explanations for why these values were
suggested in original ("sample") benchmark config. Per my understanding of the
reasons provided it makes good sense to keep these values as defaults for ML
benchmarks as well (warmup for the matters of JVM hotspot and threads for
expected typical real-life benchmark configuration).
Your suggestion to run on real multi-pc environment is particularly appreciated
because this helped to find and fix a few bugs that were rather hard to spot on
a single machine. As of now benchmarks run smoothly when executed on multiple
real hosts.
Following your other advice I also reviewed charts generated for benchmarks by
yardstick. As far as I could tell these look reasonable (as a point of
reference I also found and studied some similar charts generated for older
known good benchmarks).
-----
Summing up above, I think at this point this PR is ready to merge to master. If
you have more questions or suggestions, please let me know.
> First performance measurement of implemented ml algorithms
> ----------------------------------------------------------
>
> Key: IGNITE-6123
> URL: https://issues.apache.org/jira/browse/IGNITE-6123
> Project: Ignite
> Issue Type: Improvement
> Components: ml, yardstick
> Reporter: Yury Babak
> Assignee: Oleg Ignatenko
> Fix For: 2.4
>
>
> We want to start tracking our performance to avoid performance degradation.
> Also we need some performance comparison with other ml libs.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
