I have a Xyratex sp-2224 JBOD enclosure (dual controller) and a pair of Dell servers with SAS HBAs installed that I'm looking to utilize in a redundant NFS cluster. The hardware isn't what I'm struggling with here, its the software to do it.
I've looked into Windows Storage Spaces, and it seems like it will work using MSCS, however I'm hoping for a Linux-based solution to this problem. My Google skills are really failing me here as I've been unable to find any solid examples to help me. Most of them seem to assume a SAN backend or the like.
The end goal here is to introduce redundancy on both the disks and the NFS daemon itself. I know I can use something like Heartbeat/Pacemaker to cluster NFS, but the shared storage is a mystery to me… What filesystem(s) and/or solutions can I use to achieve something like this?
The JBOD enclosure does not do RAID, it simply presents the disks to the connected server.
Thanks in advance!
Best Answer
You're looking for: clustered storage heads connected with a dual-path SAS topology.
QuantaStor(Linux) and NexentaStor(Solaris) can do this. You can also roll-your-own with a clustering suite like RSF-1.
But really, depending on the requirement, this just adds complexity with a tiny increase in availability. The likelihood of complete head node failure is low. Things that can cause a service outage of one can definitely impact the other. Facility environment, JBOD backplane failure, power, software/OS...
You may be better served by replicating data at another tier (application?) or two parallel storage stacks and leveraging some sort of synchronous/asynchronous replication. This means two servers + storage. DRBD falls under this category, as well as some commercial solutions. See Zetavault and the offerings from SIOS.