Failover Clustering question

I've done some research on Failover Clustering, more specifically the feature offered by Windows Server 2008. I'm aware that for a two node cluster to operate there needs to be centralized storage to which both nodes are connected to.

Now the point of the failover cluster is to tolerate faults, but what doesn't make much sense is using a typical NAS, what happens if there is a fault with the NAS?

How do you suggest this be combatted, with the use of two NAS?

Why can't the nodes use the hard drives inside them?

Thanks.

 

You'll want to use centralised storage with built in redundancy. This usually involves a dual-head NAS or SAN.

 

Windows Failover Clustering is all about providing High Availability at the operating system and application layer. The assumption is that your storage layer is already properly configured and fully redundant. Failover cluster doesn't help provide storage redundancy at all, that's not it's use case.

Remember that failover clustering is not a DR or backup solution, it's a HA solution. You still need appropriate storage replication, mirroring, backups etc. as per your application requirements.

 

I've recently discovered and fallen in love with Gluster.
http://www.gluster.com/

I used to be a DRBD kind of guy, but Gluster now gets installed by preference (with the exclusion of database storage, which either gets put on DRBD still for small systems, or uses native DB replication/clustering for large systems).

 

I wasn't terribly impressed with GFS. There was a high overhead in setting up and maintaining the locking/fencing, and the throughput was pretty unimpressive. Likewise Lustre (different concept/goal to GFS) is heavy on overhead.

GlusterFS beats them both hands down. I can get a Gluster setup up and running in minutes, and it can scale in performance and size linearly with ease. Maintenance overhead is near zero, and depending how you set it up post-crash data recovery is trivial also thanks to it's design.

While Gluster isn't exactly the same design concept as GFS, they do overlap somewhat, and Gluster can make a viable target for GFS replacements.

 

Replicated storage across multiple [multi-path] Storage Arrays.

 

Agreed with everything elvis has said about gluster ... I'm about to put it out in front of one of our SANs to test it out for NFS delivery to VM. Only real concern I have is that they don't recommend using it for database applications like oracle, mysql, ms-sql ... not sure how that translates to running these things in a VM on gluster. That's easily mitigated by having database/transactional application on FC Luns, and other stuff on NFS.

If gluster supported CIFS with AD authentication out of the box ( coming soon they tell me ) I think they'd be a great replacement to the NetApp filer, which in my opinion is an average at best product for nfs/cifs sharing.

 

You can build it yourself easily. No need to use the provided GUIs if you're impatient.

The "Gluster Platform" is just a Fedora 13 build with GlusterFS, NFS and Samba all preconfigured. There's nothing stopping you building the same (or using a different distro of choice).