CPU best practise

[Ash1010]

Code compile time on vspher clients is around 10 minutes whereas on a physical box it is around 2 minutes .

The spes are below

The ESX vSphere servers are 64 bit and 8 CPUS x 2.533 GHZ

The desktops is running Intel Core 2 Quad Q9505 Processor.

Developer VMS (XP) are 2 CPUS with 4096 GB RAM

Increasing the CPU shares to high brings the compile and build time to 3.5minutes.

To rule out disk contention I moved one of the VM's to an empty LUN but got the same results , ie , 10 min and 3 min with CPU share set to high.

I was thinking of resource pooling all the developer VM's or maybe individually increasing the CPU shares on the VMS but no sure what the best option would be ? The time difference is quite big .

Any ideas ?

Thanks

Message was edited by: Ash1010

[golddiggie]

What are the actual processors inside the ESX hosts? What is the make, and model, of the host servers?

The Q9505 is a 2.83GHz processor (if you haven't OC'd it)...

Need to know the rest of the spec's such as how much total RAM is inside the host servers. What is the storage array being used for the VM's? Type of hard drive inside the array, as well as spindle speed of the drive? How is the ESX host connected to the storage array? If iSCSI, how many NICs do you have configured for the iSCSI connections (per host)? How about NIC's provided for the VM traffic?

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.

[Ash1010]

Hi

The ESX hosts are HP Proliant BL 490c G6 . 8 CPUS x 2.533 GHZ intel Xeon E5540

2 Sockets and each of them have 4 cores so 4 cores per socket . Memory is 72 Gb per host and there are 5 of these in one cluster running DRS only.

So total number of hosts are 4 and total processors are 40 with a total of 154 virtual machines only . Total memory is 360 GB

Storage system is SAN eva 6000 with 15 k disks in RAID 5 and connection to SAN is via HBA (dual pathing)

No over locking done.

[AndreTheGiant]

If compile task is only a CPU intensities operation (as usual is), then it will take less time on faster CPU.

And in this case faster CPU are usual those with more frequency.

Check CPU contention (if you change shares and it goes faster, than it may mean that you have CPU contention on your virtual environment).

Andre

[Ash1010]

Yes , increasing shares makes it faster. How do I account for cpu contention?

[AndreTheGiant]

You can check host performance monitor.

Or use esxtop (check on Internet on how to read his data).

Andre

[Ash1010]

Have checked and nothing stands out.

[dkfbp]

I would look at ESXTOP. Look at %used %rdy (take a screenshot of esxtop while a machine is compiling)

Because you see such a huge improvement by increasing the shares to high I would guess your %rdy times are high. Lets see the screenshot.

Another way to test how fast it can perform in a virtual machine is to set a CPU reserveration at 5000 mhz for one of the machines.

Best regards

Frank Brix Pedersen

blog: http://www.vfrank.org

[Ash1010]

Setting to 5000 GHZ brings it down to 3 minutes (tested on normal shares)