cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter Schuller <>
Subject Re: Defrag
Date Thu, 03 Mar 2011 08:09:32 GMT
> Potential sources of simultaneous disk write :
> - Memtable flushing
> - SSTable compaction/anticompaction
> - SSTable streaming
> - Commitlog writes (which should be on a different drive from data per
> best practices anyway..)

Honestly though, last time I measured the effects of multiple
streaming writers on modern fs:es (xfs and reiserfs at the time,
several years ago) no appreciable difference was found in practice in
terms of the resulting throughput you would get from reading the files
back. But then, if you're unlucky you might trigger some degenerate
case in allocation strategies, so who knows...

But yes, it's certainly a very valid point that concurrent streaming
I/O can be an issue. But as you say, given that there is a reasonably
small amounts of concurrent writes that (except for the commit log)
write in large chunks, it should not be a major concern.

/ Peter Schuller

View raw message