[Gluster-users] glusterfs performance

[Graeme]

Anand Avati wrote:
> Sorry for the wiki not being clear. While it is true the performance
> translators can be loaded on either client or server, the
> effectiveness is not the same. write-behind is more effective on the
> client side.

Alright, just finished a new batch of benchmarking on 1.4pre5, with
somewhat surprising results:

1.3.10: write=20.8MB/sec, read=36.1MB/sec, rewrite=12.5MB/sec
1.4pre5 (with read-ahead, write-behind on client side):
write=13.8MB/sec, read=62.9MB/sec, rewrite=10.3MB/sec

So, the read performance enhancement is very obvious with 1.4, but for
some reason my write performance is greatly reduced. I'm not sure how to
interpret this, given everyone saying writes should be much faster with
1.4 and with write-behind on the client side.

Can someone explain this behaviour, and perhaps hint how to get my write
performance back to 1.3 levels without losing the gains in read performance?

G