Shared Storage on Linux - Ideas!

[brian_31]

Greetings!

I wanted to ask this Elite group the ideas of the standards around how the Linux server should be setup around shared storage. This discussion is around clustering and should ALL LUNS be shared on clustered server except SAN boot. Please present all your ideas! Thanks for your time in advance.

Cheers!

Brian.

[Ivan Ferreira]

There are several shared storage options. ¿Can you describe your objetives, requirements and services that will be provided?

[brian_31]

Ivan,

Thanks! Here is the Scenario.

In a clustered environment (e.g., VCS) with SAN storage, how should LUNs be allocated to the servers in the cluster? Our Storage standard is that all LUNs are allocated to all servers (excluding the OS boot LUNs for the physical servers). Does this appraoch have any limitations for RHEL 5? OR are they any better alternatives?

Best Regards

Brian.

[Ivan Ferreira]

You have options liks OCFS2 and Red Hat GFS. You should have shared storage only for data that really should be shared. What is the point to have shared storage for a database that runs only on one server? If it's a file server, then you can balance the load between the cluster nodes.

[Serviceguard for Linux]

When you are talking about a cluster, all LUNs that may contain data that will be shared, should be presented to all nodes in the cluster. This is the rule-of-thumb for all HA clusters.

Also realize, except in the case of a clustered file system, or Oracle RAC, or some VERY specialized situations, a LUN is used to store the data for a single application (or application group) in the cluster and is only active on one node at a time - the node running that application.

The only exception general, if you are really paranoid, is the case where an application is only allowed to run on a subset of the nodes in the cluster, you can then set up your environment to present the LUNs to only those specific nodes of the cluster.

Finally, I believe that RHCS can support using LUN zoning as a method of fencing (data integrity) so that only the node running an app can even see that LUN.

[brian_31]

Thanks!

Is there any special config that we have to do with RHCS to acheive lun zoning? (Data integrity)?

Thanks

Brian.

[dirk dierickx]

symantec also has CFS available, which builds upon VCS to provide shared SAN storage.

another cheap way might be to provide NAS instead of SAN access, this would then be using NFS.

[Matt Palmer_2]

Hi,

if your plan is to be active/passive you could use the High Availability Extension that Novell/SuSE provide (you can get the non-suse version of this for distros like Ubuntu using Heartbeat or Pacemaker) and use the package to define resources including LUNs shared from SAN/other which are repointed to the 'other' node in the event of your defined set of failures.

I am using this in production for a High Availability MySQL installation, where I need the data migrated to the other node if the primary fails. It works really well for me. It also means that I can use some of the SAN built-in features to extend the capabilities of the storage, and hive data off using Snapclones,etc.

would be happy to share configs if it helps

hope that helps

regards

Matt

[Serviceguard for Linux]

Hi,

In answer to your RHCS question, that is in their documentation. But you should be aware that is not their first recommendation. Their normal method of fencing is to have a member of the cluster power reset the failed server. For HP servers, the use of ILO is typical.

[Alzhy · Accepted solution]

Your options for shared or "clustered" storage are plentiful on Linux. To wit:

1.) RHCS (Redhat CLuster Service) GFS
Easy to setup, support is with your RHEL subscription -- likely no issues with support as you deal with 1 vendor - Redhat.

2.) OCFS2 (Oracle Cluster FIlesystem)
Very easy to set up and install to and is now part of Oracle Linux via the Oracle Enterprise Kernel Offering. Support is via your Oracle Support subsciption. Free - generally.

3.) Veritas CLuster Server
Might be the more complicated to set up as it includes VxVM (and VxFS) and is likely the most expensive. It is however the premier clustering solution for Tier-I companies.

CLustering main ingredients include (1) the ability for storage to be immediately or always available on other nodes (2) the ability of a service to be failed over - IP, filesystems, etc to a different node and (3) the ability of more than one node to help in processing duties via shared access to clustered storage.

In my shop we use it for Shared Oracle Archiev Log access, ORACLE RAC and simple Oracle DB failover for small DBs reside on Clustered storage - vxfs, gfs or ocfs2.

Many many other useful things one can use clustered storage.