Dedicating VM Cluster to Exch 2007 and SQL

[JFWilmer]

Folks

We are getting ready to make a big push into VMWare.

We plan to build a 7 to 10 server cluster to support about 150 servers (apps servers, web, Microsoft services AD etc) general production servers.

Our plan also includes building 2 separate 3 server clusters for Exch 2007 and SQL.

As far as Exch goes each of the servers would have 3 server roles Mailbox, CA and Hub and maybe AD. Our thoughts are to put these on healthy servers with enough headroom to support HA if a server should go down. As far a SQL goes we plan on creating multiple SQL servers supporting less DBs to make their footprint smaller for DRS and HA. Both of these clusters will have RAID10 disk access.

These servers are critical to us and we just want to keep them separate from the blue collar production servers with room to grow and move around.

What are your thoughts on this approach??? Good idea??? Does it limit us in anyway??

Thanks again

[azn2kew]

Sounds like a good plan right there but what is your storage solution? FC, iSCSI or NAS? How's your disk I/O and network throughput? Its very critical to have Exchange and SQL servers have additional layer of high availability such as MSCS because HA itself only protect you from ESX host hardware failured. It doesn't protect you from services within the VM itself so you should have SQL and Exchange cluster in a box as well. There are docs from VMware how to configure them so it should be nice.

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

Regards,

Stefan Nguyen

iGeek Systems Inc.

VMware, Citrix, Microsoft Consultant

[Rodos]

I am a little confused as to your reference to clusters, may be I just need to read the post more carefully. Are you saying you are going to break your set of ESX hosts into mutliple VC clusters and then put different work loads on those different VC clusters? Depending on your overall size you may end up creating areas of isolation for spare resources and you may be better off with a large cluster and using appropriate resource pools and shares to manage the allocation and priorities. To fully understand one would want to understand the number of hosts and their size along with the size and number of your VMs. Sometimes its just better to let VMware work it out (DRS, pools, shares) than try to estimate work loads over time.

Just a thought. Maybe I miss read the direction in the post.

Considering awarding points if this is of use

[ChrisDearden]

What put me off running MSCS within an ESX cluster is the lack of vmotion support. MSCS will give you a faster failover than VMware HA in the event of a host failure , but it wont make the service any more relliable ( although any outtage of patching is greatly reduced with MSCS to the time it takes for a single failover ).

If you are going for multirole Exchange VM's I assume you'll be running them with multiple vCPU's ? in which case I've heard that running all your multi vcpu guest in their own cluster can help prevent scheduling conflists with single vcpu guests.

[Rodos]

I've heard that running all your multi vcpu guest in their own cluster can help prevent scheduling conflists with single vcpu guests.

That's a new one, having a separate cluster. It is good to keep the number of multi vCPU VMs to a minimum as that number of cores need to be available for scheduling but splitting the cluster is not going to improve this a lot. Of course there are always exceptions but you make it sound like like a general rule. If you find some multi vCPU VMs are not getting scheduled enough you should probably start by increasing its shares.

Considering awarding points if this is of use

[Rodos]

To follow up on my own post. This article looks like it is refering to the issue mentioned.

Last, you can group all your two vCPU machines together where those pesky single vCPU virtual machines won’t bother them. When a two vCPU virtual machine stops running it’ll always free up two physical CPUs. This usually means cutting up a cluster, though, so that will have also have drawbacks.

I see where its coming from. However the real problem here is that you are oversubscribed on CPUs to schedule onto. If you took the same number of hosts and then hived off a minimum of two nodes (you still want HA don't you) then you have just isolated a lot of your compute resource. You would hope that the amount of resource those mutli vCPU machines needed matched that little cluster nicely or you will have an even bigger contention problem. Sure if its a big cluster it may not be so bad, but then if it was a big cluster this is not going to be such an issue.

Would be interested to hear any peoples other real world experiences.

Considering awarding points if this is of use

[ChrisDearden]

It was mentioned by one of our Architects - I think their rationale was that the multi vcpu cluster would possibly be running a lot less CPU Overcomittment. EDIT: in addition to free resources becoming available in 2thread chunks. as mentioned above.

Personally I trying to avoid multi vcpu machines like the plauge and would rather build an exchange environment with a larger number of single role guests to scale out.

In 3.5 the co scheduling seems to have improoved considerably ( )

[azn2kew]

Read the case study on Exchange 2007 with VMware ESX 3.5

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

Regards,

Stefan Nguyen

iGeek Systems Inc.

VMware, Citrix, Microsoft Consultant