2010-05-06

MMS 2010 Labs: Powered by Hyper-V -Whatif

No - I have not gone over to the Dark Side. And forgive me for the Powershell pun, but I read an article by Ronald Beekelaar on the Microsoft Virtualization Team Blog - demonstrating the wonders of Virtualization that were used in MMS 2010 Labs: Powered by Hyper-V, System Center & HP...

The numbers are impressive - very impressive - I would suggest that you read the article.

The numbers were as follows:

  • 6 half racks - ~7 Servers per rack
  • 41 HP Proliant DL380G6 servers (Dual socket, quad-core, Nehalem Processors with SMT, 16 LPs per system) each configured with 128 GB of memory and 4x300 GB SAS drive of local storage striped
  • All networking was 1 Gb/E switched (no 10 Gb/E) and demonstrates the efficiency of Remote Desktop Protocol (RDP). Even with hundreds of labs going on simultaneously, network bandwidth was never an issue on 1 Gb/E
  • Windows Server 2008 R2 Hyper-V and System Center
  • Virtual machines were configured on average with 3-4 GB of memory each and the majority of labs used multiple VMs per lab.
  • ~40,000 Hyper-V VMs for ~80 different labs in 5 days

Ok, so I did a bit of math. I came to the conclusion that the most that they could run on 1 server at any given time would be 40 VM's per server.

Let me explain the calculation - and forgive me for my assumptions, and I not sure that they are 100% accurate but you will see where I am getting here.

My assumptions:

  • The hosts were not CPU / network / Disk constrained
  • RAM Usage on each host should not be higher than 120GB (90%) - I was being conservative.
  • The average RAM used per VM that was quoted was 3-4GB, I assumed that it was 3GB .

Using those assumptions I got to the following numbers.

120GB (Host RAM) / 3GB (VM RAM)  = 40 VM's per host.

40 (VM's) x 41 (Hosts) = 1640 Virtual machines running simultaneously

So 1640 VM's running simultaneously is a good amount of VM's, now of course not all of them run at the same time. and bringing up labs and down throughout the 5 days - I can imagine how they got to the impressive number of ~40,000 Hyper-V VMs.

But I wanted to try an see what would the comparison be with vSphere.

Again I made assumptions (in addition to the ones above) - which I think are safe and conservative.

  • Since almost all the VM's on the host are the same OS, same configuration - the amount of memory that could be reclaimed was 30%.
  • I did not calculate any further benefit from the additional shared memory after that.

Using those assumptions I got to the following numbers:

40 (VM's) x 3GB (RAM) = 120GB

30% of that that RAM is Shared - 120*70% = 36GB

36 (GB) /  3 (GB per VM) = 12

40 (VM's) + 12 (VM's - from page sharing) = 52 VM's

1640 (simultaneous VM's) / 52 (VM per host) = 31.5 (Servers)

31.5 (Servers) / 7 (Servers in a rack) = 4.5 (Racks)

10 (servers less) x $20,000 (Cost per server) = $200,000 less using vSphere

This is not taking into account:

  • 10 (servers) x 200 (watts per server) = 2000 watts
  • Additional charges for shipping costs
  • Network Components
  • etc.

Perfect example of - even if vSphere is more expensive, it will save you a hell of a lot of money!!!!!! And to quote Microsoft themselves

Still, in the end, the big question to ask yourself is the following: Is it worth all this expense for VMware, when the Microsoft solution offers a comparable or even better feature set for much less cost?  Is it worth the extra line item on the invoice, the extra line in the budget, to use VMware virtualization when it's built into Windows?  That is a question I think many customers will be asking themselves in the coming months and that is just another reason that you should start using Microsoft Virtualization solutions.

So do you know the answer???