Configuring a new cluster

[Kevin Carter_3]

We are configuring a 2 node VMS cluster using DS20E Alpha Servers. The system interconnect will be dedicated gigabit Ethernet using a cross-over cable. The storage subsystem will be a SAN with redundant controllers (2 per server). The DS20's will be configured with separate system disks (fault tolerance). Question: can the quorum disk be put on the SAN? The purpose of the cluster is to run Oracle RAC. Comments / suggestions on the proposed configuration are welcome.

[Robert_Boyd]

I'm not sure about the answer on the quorum disk, but I'm betting that you can put it on the SAN. In addition, to make it easier to administer the environment I suggest you read the section in the cluster documentation about creating a common environment in a non-shared system disk configuration. The more recent versions of the documentation make it pretty easy to figure out how to set this up by moving the right files to directories off of the system disk. You can also help performance somewhat by getting the load spread more evenly to several spindles.

Robert

[Andy Bustamante]

Question: can the quorum disk be put on the SAN?

Yes. You probably want to make sure the quorum disk has some sort of redundancy SAN level. You can't use host based volume shadowing on a quorum disk, but you can use a raid-1, raid-5 or now virtual raid to ensure availablity.

Robert makes an excellent point about relocating common files to additional disk.

If you're looking for availablity you should also consider LAN failover (and/or failSAFE IP) as well as the dedicated cluster interconnect.

Andy

[Uwe Zessin]

The quorum disk need to be on a shared bus. Most likely you don't have a shared SCSI bus, so use a disk on the SAN.

[Robert Gezelter · Accepted solution]

Kevin,

The answer is YES (been there, done that). I works quite well.

In an environment with two system disks, and a quorum disk, you want to:

- hardware mirror the quorum disk at the controller level (to deal with the contingency of drive failure while running with one of the two cluster members already down); and

- document the procedure to move the quorum disk(s) to a different SAN location in the event of a SAN problem of some sort (including a failed array or controller).

Since the quorum disk is a shared resource, it is a good place for certain information which must be used by both systems, such as
the:

- SYSUAF.DAT
- RIGHTSLIST.DAT
- PROXY files
- any similar, locally developed files

You also want to (seriously) consider getting a second LAN (even 100MB is probably more than adequate) between two nodes to prevent a $2 10BaseT, Caegory 6 cable from becomming the single point of failure (referece: "Horse show nail...")

I hope that the above is helpful.

- Bob Gezelter, http://www.rlgsc.com

[Robert Gezelter]

Kevin, and forum colleagues,

I had a bit of finger and eye trouble while typing my previous entry in this forum. The reference should read:

"Reference: The horse shoe problem ... for want of a nail..."

My apologies for the typo.

- Bob Gezelter, http://www.rlgsc.com

[comarow]

As for the quorum disk, yes, it must be on a shared disk.

However, if the disk is shared on on a busy controller, the fiber controllers internal cache can be thrashed. This will cause time outs on reading back data from hitting the quorum disk. Most of the time this will just be inconvenient, but you could get clue exits.

We have also seen in large clusters, the response from the quorum disk comes back faster than the rest of the cluster, and the other nodes clue exit.

A good idea is to increase qdiskinterval to a larger number to give the I/O a longer time to return.

Bob Comarow

[Kevin Carter_3]

The obvious answer to the next question is to â put it on the SANâ ; but Iâ m wondering is there is another solution?

Each of the DS20â s has 6 internal disks; mirroring both system disks leaves 4 disks on each node. My original thought was to have a user data disk on node A mirrored to node B. But if node A is down we are out of business. Is there another solution?

[Kevin Carter_3]

Really wish I would have turned the page in the manual 'before' posting the question about user data. The manual clearly states "Use volume shadowing to create multiple copies of all essential data disks. Place shadow set members on at least two nodes to eliminate a single point of failure." I am going for a cup of coffee...