Do you mean from client nodes to server nodes? Or are you saying
between the server nodes? Also, do share how you are measuring high IO
and what you are seeing. Some data around this behaviour would be
helpful.
On Thu, Aug 25, 2011 at 2:46 AM, Etienne Lyard <Etienne.Lyard at unige.ch>
wrote:> Hi all,
>
> we have a small gluster setup currently running on two nodes, with 2x2TB
> disks each, serving a distributed-replica volume.
> We are planning to deploy this setup over more nodes, 10 to start with,
> possibly 20 to 50 in the future.
>
> We are already noticing a significant amount of IO operations across the
two
> nodes when writting data to the filesystem, thus we were wondering: Will
the
> computational overhead induce by Gluster remain more or less constant when
> adding more nodes, or is this overhead increasing with the number of nodes
> that form a volume ?
>
> I understand that under specific conditions, a client will query all nodes
> in order to find the actual location of a specific file: this sounds like
it
> could become a problem if the number of nodes becomes large.
>
> Thanks for your insights,
>
> Cheers,
>
> Etienen
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>