MSCS & ESX 3 in production environments

[IrNico]

Are there companies which are using MSCS on ESX3 on large scale for production environments? If yes, do they use DRS?

[mreferre]

I doubt there are <many> companies using pervasively MSCS within vm's in production environments. Part of the reasons is because VMware HA addresses (not all but a large chunk of) what MSCS typically does and part of the reasons is because of the very tough deployment limitations VMware imposes. Off the top of my head one of these limitations is that VMDK files for the MSCS nodes need to be local (i.e. not shared) to specific ESX cluster nodes (unless this has changed recently... haven't checked I must admit)... thus invalidating the requirements for vmotion of the MSCS nodes and in turn invalidating the DRS requirements.

Massimo.

[TomHowarth · Accepted solution]

Yes corporations run MSCS in a virtualised environment, however it is not supported on 3.5 and only supported if the Cluster hosts VMDK's are locally stored, this fact precludes the use of DRS, HA and VMotion. as it is a requiement that all VMDKs be available to all host via shared storage.

Tom Howarth

VMware Communities User Moderator

[opbz]

Maybe not in the fully virtualized environment but I have seen a some places where they have physical to virtual clusters...Obviously active passive clusters but does provide a contingency in case a phyical box has issues...

[IrNico]

VMware HA is only a solution for hardware failures, but not for OS- or application (file)corruptions. MSCS instead gives you in this cases a lower RTO. Is this correct?

[TomHowarth]

Yes and No, MSCS has its place, but in the majority of impliementations the introduction of ESX Enterprise has removed the need for Clustering. the majority of companies do not require 9.999 or greater. therefore HA is enough. the ability to VMotion in 99% of cases give as good as or better uptime than a active pasive cluster.

MSCS brings a added complication layer the majority of companies could do without, but It does have its place, and that is in the highly redundant/highly available space. so that if a Guest crashes business continues on the second node. this however is at the detriment of losing HA/DRS and VMotion.

Tom Howarth

VMware Communities User Moderator

[IrNico]

Tom, sorry, but my English is not that good. But what do you mean with your last sentence "this however is at the detriment of losing HA/DRS and VMotion."?

[TomHowarth]

for a MSCS cluster to be supported on ESX 3.x.x the system drived must be local to the host, therefore no DRS. HA or VMotion of the guest. there is as yet no support for MSCS in 3.5 however it does work.

Tom Howarth

VMware Communities User Moderator

[mreferre]

I have had this post in draft for a few days and only this afternoon I did hit the button ..... Tom (as usual) touched on the major points already .....

http://it20.info/blogs/main/archive/2008/03/26/102.aspx

Massimo.

[dpomeroy]

Yes corporations run MSCS in a virtualised environment, however it is not supported on 3.5 and only supported if the Cluster hosts VMDK's are locally stored, this fact precludes the use of DRS, HA and VMotion. as it is a requiement that all VMDKs be available to all host via shared storage.

Tom Howarth

VMware Communities User Moderator

This is why we decided to not do MSCS in our virtual environment. If an application must be clustered, we go with physical boxes. As far as MSCS working but not being supported in 3.5, what I don't get is if a (production) application is important enough that you are going to cluster it, then why would you run it in an unsupported configuration?

Don Pomeroy

VMware Communities User Moderator

[TomHowarth]

Why thank you Massimo, where have you been lately not seen you online recently

Tom Howarth

VMware Communities User Moderator

[TomHowarth]

This is why we decided to not do MSCS in our virtual environment. If an application must be clustered, we go with physical boxes. As far as MSCS working but not being supported in 3.5, what I don't get is if a (production) application is important enough that you are going to cluster it, then why would you run it in an unsupported configuration?

Don Pomeroy

VMware Communities User Moderator

That is my point exactly and the major reason several of my customers are holding off 3.5 upgrades. these are medium SMB's that have the vast majority of their infrastructure virtualised and that includes MSCS. I assume you work in a large environment Don

Tom Howarth

VMware Communities User Moderator

[mreferre]

Tom .... on the road trying to convince the masses how bad MSCS is on VI3.... just kidding (so to speak ...)

Massimo.

[TomHowarth]

Yes I just read it, upto your usual standard. excelent.

Tom Howarth

VMware Communities User Moderator

[ncarde]

My 2 cents:

There are what I consider uncommon instances where true HA is a requirement -- particularly from a 'protection of guest memory space' perspective.

To get specific SAP has an Enqueue Replication Service (ERS) responsible for maintaining remote database state (simplified explanation) in RAM that absolutely must be deployed in a manner that guarantees that the OS's involved in delivering this are not powered off (thereby losing memory state).

VMware HA can not meet this requirement.

MSCS on VI3.5 can and it is fully supported in a multi-host configuration as long as the Cluster does not exceed 2-nodes.

So...this is not a high value cost savings opportunity but since HA/DRS are not required you can reduce some of your licensing cost and still end up deploying less Physical Hardware -- i.e. we can deliver this capability in a data center leveraging 2 VI-3 Physical Hosts (hosting two 2-node clusters) vs. funding 4 separate physical Servers.

Yes - we have to make sure the OS partition resides on a local (to each physical host) VMFS partition and that we leverage RDM's for Shared Disk (Quorom and App) but in my mind it is not as complicated as some folks make it out to be.