Reviewing our VMWare setup

[birchanger]

It's that time of year again, and this year I'm thinking about what to do with our VM infrastructure, in particular to improve performance and resilience. It was installed about four years ago and hasn't really been seriously thought about since. Most things are virtualised under vSphere 5.1 with HA configured.

 

We have a chassis containing 7 HP ProLiant BL460c G6 blades, each with 8 cores and 16GB RAM. On five of these we run:

 

• Two DCs (with DHCP, DNS, Radius, KMS, certificate services and so on). These also act as file servers (with FRSM); one hosts Impero server
  
• A host for Eclipse (Library); SpiceWorks; licence server (Autodesk, Sibelius)
  
• Symantec SEP and WSUS server
  
• Management server. On management VLAN, used for ProCurve Manager; Trapeze Ringmaster; access to switches, SANs and so on
  
• Software Deployment Server (GPO-installed software packages); SharePoint
  
• Exchange (except for Unified Messaging, which has its own blade)
  
• Finance server (PSF Financials)
  
• Print Server; including Equitrac print management
  
• MDT/WDS server
  
• LAMP server (email address mangling, Wiki)
  

 

Above list is far from comprehensive in terms of services. All the above are on Server 2008R2, except the obvious one.

 

We will need to get our CMIS setup off of Server 2003, so will probably virtualise that very soon. We also want to start playing with Server 2012R2 and System Center soon.

 

The virtualised stuff is managed by vCenter which sits on its own blade.

 

My first thoughts, without benefit of a proper capacity analysis, is that we're not short of processing grunt but the memory looks stingy, and it's rare to find a system that wasn't grateful for a few more GB. So I'll certainly add that to my shopping list.

 

I wonder about having so many services on the DCs, and am sure that shifting Impero elsewhere would be beneficial. I also think that separating out the file servers may help, and I'm considering DFS namespaces as a route towards that goal.

 

How have others gone when virtualising? More servers with one or a few services each? How has it worked?

 

Finally (for now) it seems like a waste letting vCenter have its own blade. It can't possibly need 8 cores or 16GB. What do others run it on? A dedicated but relatively puny host?

[pcstru]

How have others gone when virtualising? More servers with one or a few services each? How has it worked?

Yes - minimal services per guest server - ideally one but that rule does tend to get bent and broken when it comes to non critical services for small groups of people.

Finally (for now) it seems like a waste letting vCenter have its own blade. It can't possibly need 8 cores or 16GB. What do others run it on? A dedicated but relatively puny host?

Sort of. An old Dell 1U Xeon. Probably still overkill but it was re-tasked rather than bought for the job.

[bart21]

Virtual centre can and vmware recommend it should be a Vm guest. This is perfectly acceptable as in the event of host failure hA would still restart the VC vm.

[john]

Virtual centre can and vmware recommend it should be a Vm guest. This is perfectly acceptable as in the event of host failure hA would still restart the VC vm.

 

I would say that @Ric_ would have views over that given his morning.

 

If you have two clusters an the virtual is for managing them is on opposites that would be fine but on the same cluster its not a good idea, you don't need anything fancy for a vCentre box (a cheap desktop will do!) and your away.

Edited March 4, 2014 by john

[Ric_]

@john Is correct... running your vCenter on the cluster it is managing is horrible! The latest issue was from the host running the vCenter somehow losing its storage... the host was still up so no HA event but is mangled the dvSwitch across 4 hosts in that cluster.

 

Our typical setup is 2x DL360 (one as a DC, one for vCenter) and then everything else on HP blades. This 'just works'™.