VMWare and CPUs - Question

[vmquestions0]

I have a lab env with a Dell server that has 2 CPUs and 10Gb of RAM and ESX 3.5 install. I have some users connecting and doing some testing, nothig big just some stuff with Microsoft Office and running some other apps but all simple stuff. I have about between 8 to 10 VMs running all the time and most of the time they are idle and perhaps sometimes I can get between 2 or 4 computers with connected users (so 4 VMs being used an 4 or more up to 6 running but idle).

The performance of those machines sometimes is pretty bad and even the management of the ESX server with VC is also really slow.

I was wondering if the cpu over utilization might be the cause of the issue.

In the current situation we are running 10 virtual machine on a host which has only 2 physical processors. Not all of the VMs are being used by the users all the time but they are running.

Because of this only 2 virtual machines can execute their processes onthe physical processor and the rest of the 8 virtual machine has to wait till the processor is free.

I believe because of the reason we need to get more CPUs in order to increase the performance since it looks that the number of CPUs was not correctly choosen when the harware for the host was picked up.

Does that make sense? Also what should be the recommended number of physical CPUs if we want to run up to 12 VM's (WXP) at once also having users logged on them and working with MS office stuff (like a VDI env)?

[a_p_]

Performance issues may have a couple of reasons.

- What type of CPU' do you have (single, dual or quad core, what speed)

- What type of disks do you have (SAS, SATA)

- Do you have a RAID configuration (RAID1, 1+0, 5)

- Does the RAID controller have battery backed write cache

- How many vCPU's did you assign to the VMs

Did you already take a look at the performance charts in ESX?

André

[vmquestions0]

What type of CPU' do you have (single, dual or quad core, what speed) - Single CPU - Processor, 80556K, Xeon Woodcrest, 5130, LGA771, Burn 2

http://ark.intel.com/Product.aspx?id=27216&processor=5130&spec-codes=SLAGC

The server is a Dell PowerEdge 1950 / Service tag: 2H7QQC1

What type of disks do you have (SAS, SATA) -> The VMs are in local storage and the disk is a Serial Attached SCSI, 3, 10K, 3.5

Do you have a RAID configuration (RAID1, 1+0, 5) - Yes - RAID5

Does the RAID controller have battery backed write cache ? I am not sure what do you mean with that.

How many vCPU's did you assign to the VMs - 1 vCPU per machine

I have ran esxtop and I have attached the output. I am concerned about the %Wait time - I have noticed that when you are working with the VMs sometimes they get like froozen for miliseconds and then they work fine again, etc. I have also seen that sometimes when you are doing a logon it takes sometime while the logon it's been process, looks like they work for a while with a decent performance and then they are sluggish and go back to a decent performance etc...

Any suggestions? Do you think that an upgrade of the CPUs imrpove the system?

[RParker]

The performance of those machines sometimes is pretty bad and even the management of the ESX server with VC is also really slow.

Performance drags for DISK IO. If it's affecting the ability to manage the machine, that means something is dragging the DISK.

Are these SATA disks? that could be the culprit.

CPU shouldn't affect the ESX host that much, because the console has a higher priority than the VM's, so it should still be responsive enough within management.

[vmquestions0]

Disks are Serial Attached SCSI, 3, 10K, 3.5

VC sometimes is slow however the VMs are the ones that are slow. I am saying CPU because we only have 2 physical CPUs and they are share between the VMs so there is a %wait for them to access to the physical CPU and execute whatever process they have to do. This wait time sometimes could be longer that ohers since the previous process that is being executed from another VM is taking longer and the new process on a new VM has to wait etc...

The server has 2 single physical CPUs and 10 GB of RAM with 11 turned ON VMs at once and we can get up to a total of 26.

[a_p_]

Does the RAID controller have battery backed write cache ? I am not sure what do you mean with that.

According to the part list of your server you have a PERC5i RAID controller with 256 MB cache. Without a "RAID battery" (that's how DELL calls it in the system manual) the cache is usually set as read-only cache. To enable part of the 256 MB cache as write-cache it might be neccessary to add this RAID battery. ESX does not do write caching for disks, this must be done by the controller.

André