Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I will be looking to buy a new server soon, and would like to consolidate servers into one VM server. I will have equivalent infrastructure offset, so this isn't a case of having all my eggs in one basket.

I have a Terminal Server that is running an application that is very CPU-constrained (RAM, disk I/O and net I/O are light) running on a 2x Xeon E5405 box. I also have a Domino server running on another box (Prestonia Xeon, old), an AD Domain Controller (no FSMO roles, a member of a global tree, roughly same vintage), a lightly used file server (another Xeon of the same vintage), and a Linux server running on fairly weak hardware.

I plan to buy a server with several CPUs, and then assign all but one or two to the TS VM (which will have virtual SMP). The remaining VMs will not have virtual SMP and can be scheduled freely among the 1 or 2 CPUs remaining.

I've heard that running a VM with many virtual CPUs does not perform well -- is there any truth to this?

I will also need a backup solution that can talk to an LTO3 SCSI tape drive and be able to back up VMs so that DR would be as simple as setting up another VMWare box and restoring to it. Any suggestions?

Any other hardware or software pitfalls to be aware of?

share|improve this question
    
We ended up giving up on the idea of virtualising a heavily-used TS (performace improves going to 2 vCPUs, but by about 4 it actually drops when more vCPUs are added, so I am told), and are going to install TS onto a physical server. –  John X Citizen May 20 '11 at 8:01

2 Answers 2

up vote 1 down vote accepted

As I understand it, when a VM needs some CPU time, the scheduler only gives it CPU time when there are sufficient free cores on the host equal to that VMs vCPU count. So if you have a 4 vCPU VM, you only get CPU time if 4 cores on your host are available at that moment in time. So VMs with fewer cores have a greater change of being given CPU time.

So I guess there are several solutions to this:

  • Increase the number of total cores in your host so that the odds of having enough cores free is higher. So spec a 2 x 6-core server, or a 4 x 4-core rather than a 2 x 4-core.
  • Dedicate some cores on the host entirely to a particular VM (not sure this is possible).
  • Decrease the vCPU count on your VM (and run multiple VM instances if the workload can scale horizontally).
  • Buy hosts with high core clock speeds to do more with fewer cores.

For terminal services, my gut feeling would be that deploying multiple virtual TS instances is your best bet, with TS load balancing.

share|improve this answer
    
It is possible to dedicate cores to a VM in ESX/ESXi. Looki for CPU affinity options under advanced config for the VM. –  Jed Daniels Apr 18 '11 at 5:40
    
@Jed I thought about affinity also but saw this link vmware.com/pdf/vmware_drs_wp.pdf It's more for preferred CPU, not dedicated CPU –  ST8Z6FR57ABE6A8RE9UF Apr 18 '11 at 5:52
    
@ST8Z6FR57ABE6A8RE9UF I'm not sure that I understand how DRS relates to this scenario. DRS is a different feature in VMware. It is certainly possible to tell a particular VM to only use a particular CPU or CPUs, and then you simply tell to others to use the other CPUs. –  Jed Daniels Apr 18 '11 at 21:10
    
Thanks for the replies -- more reading has indicating that ESXi has much more relaxed co-scheduling -- the lagging vCPUs are allowed some skew before they must catch up and maybe the leading ones don't have to run until they do? –  John X Citizen Apr 20 '11 at 4:49
    
Sorry, <enter> adds the comment. If I run several TS instances, the ERP application will have to access its data files across the virtual network. This app using file section locking instead of a real database server; we have seen running this on a network causes corruption. –  John X Citizen Apr 20 '11 at 4:51

I can reply to backup part of your question. I just finished evaluation of 4 backup solution for VMware for our business and here they are:

  • Veeam Backup & Replication
  • PHD Backup
  • Quest vRanger
  • online backup based on Asigra

These are only more or less mature solutions suitable for production use I could find. I personally liked Veeam (price/value wise) but if you require backing up to tape PHD may be a better fit. The last one is what we are ended up with, but it is NOT designed for tape backups so it may be of no value for you.

Hope this helps.

share|improve this answer
    
How do these compare to vmwares data recovery? –  3molo Apr 18 '11 at 6:42
    
All the consultants are recommending Veeam as well. Might use Veeam to back up to NAS, and have BackupExec installed into one of the VMs and back up to tape from there. –  John X Citizen Apr 20 '11 at 4:52

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.