dchristle commented on pull request #10847:
URL: https://github.com/apache/kafka/pull/10847#issuecomment-858804912


   @ijuma 
   
   > This is a good change, but can we please quality the perf improvements 
claim? My understanding is that only applies to certain compression levels and 
Kafka currently always picks a specific one. @dongjinleekr is working on making 
that configurable via a separate KIP.
   
   It is true that the most recent performance improvements I quoted (for 
`1.5.0`) appear only in mid-range compression levels. I did not highlight it in 
my description, but besides bug fixes, the earlier releases quote consistent 
perf improvements:
   
   - `1.4.4`: ~10% decompression bump, no level-dependence quoted
   - `1.4.5`: 5-10% decompression improvement in `x86_64` architecture, +15-50% 
in various `arm` processors
   - `1.4.7`: Improved `--long` mode compression ratio at high levels, 5-30% 
decompression improvement for blocks < 32kB
   - `1.4.9`: 2x faster `--long` mode compression speed
   
   > Also, why are we listing versions in the PR description that are not 
relevant to this upgrade?
   
   I tried to follow a previous `zstd-jni` PR's convention here: 
https://github.com/apache/kafka/pull/10285 . I think it gives context on the 
magnitude of the upgrade, but I can change the commit message/PR title to 
remove the existing version reference if you like.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


Reply via email to