Setting up a Simple 2-Node Itanium VMSCluster

[Keith Orawiec]

I have been out hands-on work for over a decade and the last time I setup a VMSCluster was in the early 1990's. My knowledge is old and dated and downsizing has eliminated anyone else here with more currect knowledge.

Anyway, I am going to be setting up a 2-Server System with an external disk array. Basics include:
Two rx2620 servers each with one local disk
One, shared MSA1000 SAN array with 5 disks in it

This will be for an online process control application. The developed executables and data will reside on the disks in the MSA1000.

One machine will server as the "primary" online machine for control and the other machine will be "looking at" the same apps and data, in standby mode. It is a very striaght-forward approach and there really aren't any interactive users to be concerned with. The general idea here is that one machine could be literally turned off and the other can be controlling the process.

The end user has requested that the 5 disks in the SAN array be configured in a RAID 1+0 configuration. I have no emotional attachment any particular RAID configuration so I plan to go this way.

Now finally to my questions:
1) In the past, I seem to remember that the quorum disk was not supposed to be a member of a shadow set. Is the same recommendation true with a quorum disk being a member of a RAID array? I was thinking that I would RAID 4 of the 5 disks and leave the 5th as the quorum disk and a place to dump image savesets for each of the server's system disks. However, if I can have the quorum disk f part of the RAID set, it appears that I can have a little more effective and shared disk space.

2) Since these machines will be setup to run the applications without the other are there any other gotchas out there that I need to be wary of?

Thanks for any and all responses.

Keith

[Jan van den Ende]

Keith,

to begin: Welcome to the VMS forum.

My personal approach would begin with NOT having two system disks, but one SAN system disk with two roots.

And please do realise, that if you arrange your SAN _drives_ as (any) RAID, you can present chunks of that as different _DISKS_ to VMS.
For a two-node cluster to be able to continue on any single node, you DO need a quorum disk that is directly attached to both nodes, so that MUST be on the SAN in your config (and it can not be Host-based shadowed).
Your local disks are probably best used as DUMP devices (look up DOSD in the System management Manual). The saveset storage function does not look unreasonable.

This should do for a start.
More detailed info/design wishes will probably get more detailed answers from various participants,
I expect the common denominator to be generally aligned, with various personal preferenced details to pick your favorite taste.

Success!

Proost.

Have one on me.

jpe

[Robert Gezelter]

Keith,

I concur with Jan.

You will need a quorum disk. The restriction is on Host Based Volume Shadowing, not on their being a RAID array.

I agree with Jan, one system disk, possibly RAID/Host Based Shadowed, and multiple different roots.

You will need to setup a reliable way for the surviving machine to grab (and retain control) of the application. You will also need a way to voluntarily relinquish control of the application to the other cluster member short of halting a processor. Controlled handoffs are far, far safer. This can be done using the Lock Manager, but exactly how you implement it depends on how your application is structured.

I hope that the above is helpful.

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

[Andy Bustamante]

A quorum disk can be a member of the raid set, but as a rule you don't want high I/O from applications on the quorum disk. This controller can create logicals devices from all 5 disks. See http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00709035/c00709035.pdf I'd create a data disk, a small quorum disk, and allocate enough "spare" space to run with a failed disk.

Andy

[Keith Orawiec]

Thanks for the replies. So I can certainly make the quorum disk part of the RAID array. That was my main question. Beyond that, I am currently thinking that I will create three logical disks from the 5 physical disks primarily due to the design of our applications.

I guess I haven't thought through how to deal with the system disks just yet. I need to think on that. In the "olden days" I seem to remember that you didn't want the pagefile on a shadowed disk. Is this still true or have things progress since those days?

Just as a point of interest - "failover" of our applications from one machine to the other is indeed accomplished by using the distributed lock manager. The first subroutine call in every app essentially goes out and tries to get a shared lock that we create. This lock is controlled by a local task that is also looking for another lock managed by a similar task on the other node. The obvious way for the lock on the remote machine would be for it to be shutdown but we also have custom code to decide when to free the lock and "failover" based on a custom set of criteria design for this specific system. We also have code in there to prevent failover if there is an external failure to avoid the case where you just failover and back cyclicly. It's a pretty simple yet effective scheme. Unfortunately, the cluster license is required and it makes this costly. The upside is that we can achieve near 100% availability even when there is a failure. Like I said - just a point of interest for you guys.

Thanks for the replies. Once my machines arrive, I might indeed become a bit of a pest. My old brain is having trouble firing up. This will be fun!

[Ian Miller.]

you can have pagefiles on shadowed disks or in your case controller RAIDed disks.

Have a think though various failure scenarious and how you recover e.g failre of a disk, failure of another disk after 1 has failed. Failure of a system (crash or hardware problem), corruption/deletion of a file or files etc. This way you validate your system design.

[Ian Miller.]

example. You have your five disks in a single raid 5 set which is then partioned to present various drives. One disk fails - you are ok but make sure you have some way of detecting the failed disk. Two disks fail - your RAID5 set is dead, all your disks are dead and both systems are dead.