Question about RDM in ESX 3.5

[stoutat]

Here's my scenario.

Currently I host our SAN connections via 2 servers using MSCS. The shares are knows as our "sancluster". I want to virtualize this task. I'm thinking that if I present those LUNS via aliases and zoning in my Brocade software to the ESX servers and do RDM that should work. The question is - do I stay with MSCS (soon to be supported) or go with a single host and rely on HA/Vmotion (which gives me a minimal amount of downtime in the event of a failure).

Maybe I'm completely off-base and this doesn't make any sense at all.

I admit I'm a bit "green" on fiber zoning and VMWare in general.

Thanks!!!

[ctfoster]

No, not off base. The first point is that MSCS and HA are not directly comparable technologies so if you are going down the HA route you will lose some functionality. As you know MSCS has a layer of software that monitors the state of the application services running as part of the cluster which incorporates a layer of logical which can fail-over software suites independant of any issues with underlying OS or host. With HA things are much simplier. It's main function is to restart VM's should an ESX fail. Your Exchange system could be dead in the water and your users coming after you with meat cleavers - buts thats no concern to HA so long as the host is pinging.

On a positive note MSCS is often perceived as maintaining connection state during the move of the 'virtual machine' from one side of the cluster to the other which is not the case. In that respect its much like HA, the only difference being the restart requires an application start and roll-forward rather than a complete OS restart. However VM's reboot so fast anyway I often find that by the time the MSCS has 'remastered' the SCSI LUN and started the app theres not much in it.

You dont say what appplication you are supporting and how critical they are. This could be the deciding factor.

[stoutat]

The SANcluster as we call it is used for file sharing/user

shares/profiles/print queues - pretty much every important aspect of the

business.

[ctfoster]

Since its not running a cluster aware appplication such as SQL Server or Exchange - unless you include the print spooler, a lot of the advantages of running MSCS are lost. I think HA would give you almost as good a solution as MSCS. It would cheaper as well.

[stoutat]

Thanks for the help so far. I'm still leaning towards clustering. If I

"merge" a new VM in with my physical machines I the cluster, I shouldn't

have to redo any shares or login scripts, then I could eventually phase

out the physical machines. Is this a sound theory?

[ctfoster · Accepted solution]

Yes. If you add a new vm node into your cluster using VDM storage you wouldn't have to redo anything - in the same way you wouldn't have to change anything if you added a new physical node. You could then evict the other physical nodes to be left with a 'one-legged' cluster. Good luck.

[smartloom]

Got update from vmware support. Fix is not in 3.5 update1. Development is trying to debug the issue from Kernel. They will also try to simulate the issue on AMD chipset as they haven't done that. So far they have simulated the issue on intel chip sets.

[stoutat]

The fix for what? MSCS?

[smartloom]

Posted in wrong section meant to post in Windows2003 takes 6min if I restart from OS on ESX 3.5 topic... Any how About MSCS as per vmaware fix will be in Update1.