Server performance multiple VM vs 1 native OS

[SimpleSi]

Has anyone done (or got link to a reliable source - not just used Google ) any real performance/issues test on mutiple VMs doing different roles e.g a single W2k3 machine running file/print/ad/dns/dhcp) over say the same physical machine running one one of more VMs for different roles (e.g run the ad/dns/dhcp on one VM and the file/print server on another).

CAn you actually gain real world performance or are you always going to get a performance hit vs benefits of quick diaster recovery/ease of maintenance?

Si

[SYNACK]

Just to point out a bit of a flaw, 2k3 was rubbish with virtualisation due to the way the kernel was put together, although they made quite a few improvements in SP2 (hence why Hyper-V will only be happy with SP2 or above hosts) it is still junk in comparison to 2008 or 2008 R2. VMware and the others made a lot of little hacks to get 2k3's performance better under a hypervisor but it is still junk in comparison.

Given this if you are forced to use 2003 there will definatly be a nasty performance hit from virtualising. Virtualising itself with any system will add a bit of overhead but how much depends on both the hypervisor (Hyper-V, ESX, Xen) and the hosted OSs. Virtualisation will never get you more than a local install but it will get you better utilisation of the avalible hardware as for the most part decent hardware is not taxed by a few simple roles like AD, DNS and the like. This means that if you have robust hardware despite the loss of a bit of power you gain a stack of other abilities that make it worthwhile and with performance to spare thanks to more powerful modern hardware you probably won't notice anyway.

Edit: even with 2003 I'd be tempted to dump print onto a seporate box thanks to its messy nature.

[tmcd35]

It sounds like a flawed comparison. If you were to spec up 1 server for print/dns/dhcp/file/ad then it'd be a lower spec then spec'ing to run 5 VM's to do the job individually.

If you ran the 1 server on the later spec it should positively fly. On the other hand trying to run the VM's on the former spec should crawl because you are in effect funning 6 copies of the OS on one machine.

If you spec the host correctly you should seeno real difference in performance between running Vm's to running the services on one lower spec'd single server.

I'm sure there are speed comparisons out there somewhere but modern hypervisors run at such a low level that when you look at hosts with 12-cores and 48gb ram, your not going to see the difference between that and 6 dual core, 6gb physical servers.

Actually, thinking about it, your biggest performance worries are going to be with hard drive and network bottlenecks with VM's more than processor/ram bottlenecks - assuming a correctly spec'd server.

[SimpleSi]

The w2k3 thing can be changed to any WinOS you like
The comparison is meant to be on same hardware otherwise its not a comparision of OS methods

I was just wondering if they was any possible true performance gain in running multiple OSs in VMs running on a bare metal hypervisor thingy vs sticking all roles into one WinOS running natively

Si