CLVM clarification - general.
 

Hi...

I've been reading up on Linux clustering and successfully set up a load-balancing cluster in a lab environment. I'm going to tear down that setup now and I want to learn about storage clusters.

After reviewing some of the clustering documentation on the Red Hat site, I have some confusion about storage clusters and thought I might be able to get some general questions answered here.

In the Red Hat diagrams, they model a storage cluster in a manner that looks similar to several servers accessing a SAN. In my lab, I want to take a different approach.

For my purposes, I have 3 Linux servers, each system has two 40GB hard drives. My goal is to load the Linux OS on the first drive of each of the servers, and to allocate the second drive in each server to be used as storage in the Linux storage cluster.

So here are a few questions I have...

1. First... is this approach possible?

2. If so, when I create the filesystem on the secondary drive of each server, do I use GFS?

3. Are you LVM2 and CLVM mutually exclusive? If so, do I have to use CLVM to do what I'm trying to do?

Thanks in advance.

Regards,

Brett

1. not really, since clustered storage needs to be shared storage which is accessible by all nodes. you can setup some scripts (rsync?) to keep the data close to identical between each node but it really isnt clustered storage in this case but more like a distributed web cluster where each has it's own local copy of the web pages and a load-balancer round-robins (or other method of distribution) the workload between the nodes

2. you are welcome to use GFS with the lock_nolock dlm for a single instance (single node access)

3. yes, only one or the other can be in use at a time but you can switch between them (so you can use LVM for local LVM managment and then switch back to CLVM when finished). CLVM piggy-backs off of LVM's technology and features. it provides the benefit of updating the other nodes LVM config through cluster communication so that they all maintain a consistent view of the shared storage.


what i did since i didnt have a SAN at my disposal either when i was "playing" with this is configure a system to be an iscsi-target so that the nodes in the cluster could access a single storage LUN using the iscsi-initiator utils. this should be a system distinct from the 3 nodes so that you are able to use your fencing without worrying about fencing the iscsi-target node. This seemed to work well.


hope this helps.

You might also want to look at PVFS and Lustre as they're cluster filesystems that are designed for use more in high performance environments. I've never used either, but I'd like to try them out sometime...