[Gluster-devel] production use?

[Jerker Nyberg]

Hi,

I've been running GlusterFS on a small cluster (5 old nodes) for a couple 
of months and things seem to work fine. My department is planning to 
upgrade the storage subsystem for their clusters in a few months, 
currently the clusters (200+100+100 nodes) are using NFS-servers with 27 
TByte storage. GlusterFS is one of the options. (Lustre is one of the 
others and so is adding a couple of normal NFS-servers.)

Is it recommended to try using GlusterFS for production use today (or
probably early 2008)?

What is the best practice for a storage cluster running GlusterFS? I would 
guess the raw storage would be in the order of 100 to 480 TBytes (up to 
two racks of storage nodes.)

How does GlusterFS handle biterrors? Have anyone tried to run the 
GlusterFS brick on Solaris/FreeBSD with ZFS? Are there any other plans for 
checksums?

It is possible for GlusterFS to use the local disk on a client as a cache? 
I currently see no such need, but I got the question.

One of the older machines is a SunFire15k running Solaris. But it is still 
the one with the most cores and memory. Is it possible now (or in the 
future) for this machine to access GlusterFS directly as a client without 
reexporting as NFS?

Regards,
Jerker Nyberg, Uppsala.
Computer Science Student at Uppsala University.