[Gluster-devel] glusterfs system design

[Kevin Van Workum]

I know this is a broad question, but...

I now have the need for a SAN rather than NFS servers. I've been
playing with glusterfs for a few days and like what I see so for. My
problem is I don't want to spend weeks redoing what I'm sure some of
you have already done, i.e. optimizing the myriad of
configuration/design possibilities provided by glusterfs for my
particular situation. My hope is that by describing my needs, some of
you might be able to provide some good suggestions on starting points.

So here's my use case. I have a very broad range of user applications
and new ones being added everyday. Some users are sophisticated and
can tailor their applications to use the file system optimally with
input from me, others are not and do very inefficient things. I need
to be able to service them all as best as possible. And therefore need
good performance on the broadest range of file server scenarios. I
currently have O(100) client nodes and need room for O(1000) in the
near future. I have Gb ethernet and do not plan on changing that.
Though I am open to the idea of a separate network for storage.

Here's what I want. I require high-availability (afr?), better than
NFS performance, scalability to O(1000) clients, and the ability to
dynamically increase/decrease my storage capacity.

I'm pretty sure glusterfs can provide that, I'm just not sure how I
should start or what are the most import factors to consider. Rather
than tainting the discussion with specific questions, I'll just ask
you for your thoughts.

Thanks for your help.

Kevin