VMware Cloud Community
Roberto_B
Contributor
Contributor
Jump to solution

Trying to get a general grasp on this.

Hello

I am trying ot get a general idea on a project I am doing. I am sort of new to VM and I have only done 4 VM servers. This project is alittle more complicated.

What we are trying to do is shrink our NDC to VM ware. What my work partner is trying to do is convert existing servers to VM images. We have 2 Dell R900 servers with dual Quad core cpu's. These server will be the host servers but the VM's will be stored on a EMC SAN. Now what I am trying to grasp here is the servers he wants to convert are dual processor servers but only wants to issue 1 core to them. Now the single cores are about the same speed or close to it as each existing processor in the old server we want to convert. Should he issue the same amount of cores as processor the server had(both CPUS were being used 40% to 70%)? If we stick only issueing 1 core will we see a increase in latency or a peaked CPU? Will the EMC box make up for the loss of the CPU/core?

Next is he wants to convert DC's. I am afraid to do this as I was told before never convert a DC to do a fresh install. is it smart to put DC's on a VM image? Anybody do this yet? Each of our DC's host a global catalog.

0 Kudos
1 Solution

Accepted Solutions
Dave_Mishchenko
Immortal
Immortal
Jump to solution

I'd stick with setting up a new VM and promoting it to a DC over running a P2V on a DC, but if you do

1) you'll have to do a cold clone

2) review your backup strategy as snapshots are not supported by MS for backing up DCs as a restore can mess up DC synchronization.

Also if you do P2V servers from multi to single CPU, make sure to upgrade the HAL to uniprocessor to avoid performance issues with the VM. http://www.vmware.com/community/thread.jspa?threadID=84611&messageID=646375#646375

View solution in original post

0 Kudos
5 Replies
jayolsen
Expert
Expert
Jump to solution

As far as the DCs, it is far easier to just build a new one on the VM hosts and transfer the roles over to it. You can run a cold convert and it should work fine, but most people will tell you just to stand up a new one to reduce any risks for corrupting AD. We have several DCs running as VMs without any issues.

In general you only want to give your VM one vCPU unless an application specifically needs it. You could see more latency with two vCPUs.

ejward
Expert
Expert
Jump to solution

With virtualization, you've got to throw out normal ideas about CPU's. many times a physical box with 2 physical CPUs or cores will perform better as a VM with just a single virtual CPU. It depends. If you are taking that physical box and putting it on an ESX host by itself or with just 1 or 2 other VMs, chances are that it won't perform better. The more VMs you put on a host, the more chance you have that a single virtual CPU VM will perform better than a multi-CPU VM will. Huh? remember that part about throwing out normal ideas?

In the virtual world, if a VM has 2 CPUs and the application is multi-threaded, when 2 threads want to process, the VM has to wait until 2 physical CPUs are available at the exact same time before the VM can process what it needs to. If you have a lot of VMs on the host, you can get a lot of CPU ready on the VM. This means that the VM is ready to process but, the ESX host and the physical CPUs are not. With a single CPU. If you are using 2-CPU quad-core servers for your host then, you've got 8 CPUs and it might not be an issue. If you create the VM with just a single virtual CPU then, it only has to wait for ANY of the physical CPUs to become ready. If you've only got a few VMs you won't see anything. If you're squeezing 20 to 30 VMs on a host you'll notice. I had one VM that was running slow. I took away one of the CPUs and it ran faster. go figure.

As far as DC's go ... I've done P2V's on our 6 DC's for disaster recovery with no real issues. as long as all the clocks are in sync and you have the ability to force replication when the come on-line, i've never really seen a problem. It's just another server.

dmorgan
Hot Shot
Hot Shot
Jump to solution

The other poster is correct, unless you are running a specific app that utilizes more than one processor/core, you want to stick with a single vcpu. The reason being, you may receive more latency by having two vcpu's, due to the fact that you will need time-slices to be available on two physical proc's in order to process, as opposed to a time-slice on a single physical cpu prior to processing. You can always assign a single vcpu, but give it a higher reserved CPU time, thus giving it more resources while still having a single vcpu. Make sense?

If you found this or any other post helpful please consider the use of the Helpfull/Correct buttons to award points

If you found this or any other post helpful please consider the use of the Helpfull/Correct buttons to award points
0 Kudos
ejward
Expert
Expert
Jump to solution

The other thing to consider is that hopefully any physical machine you are virtualizing is older than the ESX host you are putting it on. So a physical server with 2x500Mhz CPUs is going to run better virtualized with a single CPU when that CPU is 2Ghz.

0 Kudos
Dave_Mishchenko
Immortal
Immortal
Jump to solution

I'd stick with setting up a new VM and promoting it to a DC over running a P2V on a DC, but if you do

1) you'll have to do a cold clone

2) review your backup strategy as snapshots are not supported by MS for backing up DCs as a restore can mess up DC synchronization.

Also if you do P2V servers from multi to single CPU, make sure to upgrade the HAL to uniprocessor to avoid performance issues with the VM. http://www.vmware.com/community/thread.jspa?threadID=84611&messageID=646375#646375

0 Kudos