Hi guys,

totally OT but....

My new job has a vmware farm of about 20 virtual servers on 3 physical servers.

My question is...wouldn't it be better to decrease the number of VMs...

for example they have an AD server and seperate print server. this makes sense in a physical world because of load balancing but surely not when the servers are all effectively on the same hardware resources...in fact the VM overhead must cause a negative effect.

am I right in my thinking or not? if not why not?

Cheers
Jamie

Yes and no.

The simplest use case would be when you need to separate the services to different servers for whatever reason, then they're already sandboxed for transport.

This is also great for load balancing/maximizing the server. So you can pick and choose which services to run on said server (which ones tax the CPU, network and disk loads).

All for a minimal overhead (relatively speaking).

Hi Jamie,

i work on lots of large vm environments and we regularly scale out lots of vm's on a small number of hosts. 50:1 in some cases!

Using Virtual Center, look at the cpu utilisation on the host servers and theyre probably still really low. The overhead of virtualisation is pretty low and is offset by the low utilisation vm's.

The benefit of keeping them seperate is the same as in the physical world, reduced risk in the event of a single vm outage.

Reducing numbers of vm's would simplify management and potentially reduce licensing costs. The os license is probably datacentre edition so 1 license for unlimited vm's.

At the end of the day, it's all about best utilisation of cpu and memory resources. Check virtual center and make sure cpu and memory are below 2/3 in a 3 host farm.

not sure if I answered the question, but hope it helps!

Steve

thanks guys...both your answers were great.

I'm thinking of reducing the numbers for some of the reasons you listed Steve, mainly for easier admin ,at the moment I have to guess which servers are doing what as the naming conventions are ...strange...

also whilst spreading the services across the servers reduces the risk of total failure if one vm goes down it must also increase the likelyhood of a single vm failing...(if theres 10 vms instead of 1 then surely the chances of a vm failure is tenfold)

plus if a standard 2003 server install takes up 3GB (guessing) then thats quite a large amount (60gb for 20 vms). not too much of a problem with big HDDs nowadays but then that data also has to be backed up...

and in case of a hardware failure I'd have much more work to recover all the VMs.

I think the cons outway the pros of multiple vms.

Why on earth does someone who doesn't understand virtualisation get put in charge of the virtualisation in a company?

Systems don't randomly fall down. They fall down because some hardware dies out, or because it is attacked and compromised. Other than that, they won't just "stop working" by magic. So your "tenfold of vm failure" argument is just absolute crap.

You should tell us what kind of hardware you're running on. Depending on what kind of usage the server will be on, they should be using different profiles. Get your company to send you on a VMWare course or training (whatever they're called these days).

Also, please don't take management and sandboxing decisions based on your ability to track usernames and passwords or system functionality. Create a spreadsheet that will help you retain that kind of information.

/me facepalms thinking about the future of your company.

Heh that's true... that kind of multiplication is usually only applied when calculating RAID's disk failure chance. Then again, I don't know how much you leave things out to chance when you deal with Windows VMs on a VMWare machine (I assume ESXi?).

I used debian minimal (net?) install back then, and each VM started out at around 100MB base system... with usually 20-50MB on top of them running the compartmentalized service.

You can also get considerable saving if all of the VM are based on the same disk image...