eeroel commented on issue #38664:
URL: https://github.com/apache/arrow/issues/38664#issuecomment-1811938124

   > I mean that maybe some flags in 
https://docs.aws.amazon.com/cli/latest/userguide/cli-configure-envvars.html 
helps, but I didn't test them. S3FS in arrow fs just wraps the S3 SDK, maybe we 
can find some existing solutions in S3 and port from them?
   
   Right, I couldn't find any AWS flag that would help here. There is also this 
issue about switching to the S3CrtClient in AWS SDK: 
https://github.com/apache/arrow/issues/18923
   
   I actually tried the CRT client with Arrow, it's a drop-in replacement, and 
in my limited test cases it does give higher throughput than the old S3Client. 
Not sure what the main sources of performance boost are, but it does handle the 
IP address discovery in some clever way and automatically splits GET requests 
into smaller ranges and parallelizes. The main problem I've noticed is that it 
actually does a HEAD request for each byte-range GET, in order to validate the 
range, so this would be an issue for low-latency use cases (I've been hoping 
that HEAD requests could be eliminated 😅).


-- 
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.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to