[Gluster-users] gluster stripe
Jeff Darcy
jdarcy at redhat.com
Tue Jun 29 12:45:45 UTC 2010
On 06/29/2010 05:13 AM, Nick Birkett wrote:
> Im using IOZONE. Each server has 16Gbytes mem and I'm running 32Gbyte
> files with iozone.
> The Glusterfs file system is 133Gbytes in total. So yes it could be
> cache effect since the
> files are divided by 8 on a stripe. The local storage on each server is
> just a single local disk
> and quite slow (100Mbytes/s read/write).
> The network is DDR Infiniband so a bit overkill for the slow disks.
OK, so it's not the network. If you're doing large (e.g. 1MB) writes on
the client, with the io-threads translator and multiple iozone threads,
it seems like you should be able to get much better than 100MB/s for
write. I think seeing your volfiles - especially on the client - and
iozone command would help to identify what that's not the case.
More information about the Gluster-users
mailing list