[Gluster-devel] Unify read balancing

[Jake Maul]

Greetings all,

I'm looking at rolling out a Unify+AFR cluster, as a content backend
for a few websites. One of my concerns is how well GlusterFS is going
to handle balancing reads between the content servers. The read access
will be heavily front-loaded during a file's lifespan (that is, lots
of reads the first day, considerably less the 2nd day, etc).

At first I got the impression that Unify (or, more appropriately, the
ALU scheduler) would actually move files between bricks to balance out
the read load, but further reading indicates it just alters the
creation location of *new* files.

Has anyone had any trouble with Unify/ALU inefficiently balancing read
loads between bricks? If so, is there any good way to 'redistribute'
things?

If one were to add a server to an existing Unify cluster, is there any
good way to redistribute the *existing* files to it? If not, I would
expect that virtually all new files would end up on the new server
until it filled up. This wouldn't be too hot for performance in a
situation where new files are much more popular than older ones...

Many thanks,
Jake