Performance issue on AMD QuadCore CPUs

[jbauriede]

Hi all,

in our environment we observe very different performance on esx host with quad core AMD cpu an dual core ADM cpu.

The performance on quad core cpu is much slower than on dualcore cpu. The hosted application is an ODBC application connected to an MS SQL Server 2005.

This problem is reproduceable. Can anybody confirm this fact?

regards

Jürgen

[Chuck8773]

As this doesn't make sense, I would first strive to verify that the hosts are identical with the CPU difference being the only difference.

Are the server models the same?

Check the bios on both hosts.

Are all nics set to the same duplex/speed setting?

What kind of performance degradation do you see?

We went through a similar upgrade. From dual to quad and saw no change in performance of individual VM's. The host was just able to support more.

Are the cpu's the same Ghz? The quad cores are sometimes slower individually.

Charles Killmer, VCP

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[AntonVZhbankov]

How many vCPUs are in VM?

How many VMs are on the same host?

---

VMware vExpert '2009

http://blog.vadmin.ru

[jbauriede]

Hallo Charles,

yes I've checked the hosts more than twice. The hosts are identical - only the cpu's are different, the same san, the same network.

we checked all known performance issues - cpu counters, network errors, bus resets, irq sharing and so on.

At first we run the application on our quadcores. At next we make some adaptations like cpu pinning and reservation. At this time the repeat time of the application was nearly the same.

Now we shut down the VM's and migrate to the dualcore cpus. these hosts have identical HBA's and nic's. The next step was to power on the VM's and make the same tests.

I was very confused that the response time was nearly cut into halves.

regards

Jürgen

[jbauriede]

Hi Anton,

the application contains two VM's, one application VM with 2 vCPU's and one database hosted on a 4 vCPU VM.

During these tests, each VM run alone on a esx host.

regards

Jürgen

[AndreTheGiant]

So you have 3 ESX with each 1 VM?

You have performance issue only on applications server?

What are the hosts CPU configuration? (CPU type and frequency)

Andre

[jbauriede]

Hello Andre,

I have any HP DL585 G5, 4x AMD QuadCore Opteron 2,2GHz, 128 GB RAM and any HP DL585 G2, 4x AMD DualCore Opteron 2,6 GHz, 128 GB RAM. I tested the application in the following szenario:

applications server on HP DL585 G5 <-> database server on HP DL585 G5 --> response time approximately 15 sec.

applications server on HP DL585 G5 <-> database server on HP DL585 G2 --> response time approximately 12 sec.

applications server on HP DL585 G2 <-> database server on HP DL585 G5 --> response time approximately 12 sec.

applications server on HP DL585 G2 <-> database server on HP DL585 G2 --> response time approximately 7 sec. !!!

The performance issue is not only on the applications server.

Jürgen

[AndreTheGiant]

As you see the frequency are not the same.

If you have an application CPU intensive than a CPU with 2,6 GHz could be better than a new 2,2 GHz (using the same number of core on both system).

Also the memory access could be better (using the same bus frequency) on a dual core, cause each core has a dedicated cache.

Andre

[jbauriede]

Hallo Andre,

I agree with you, but the VM's uses approximately 8 - 25% of cpu resources - I think this is not a high cpu usage.

VMWare says " ... the question has been referred to a research engineer for further analysis. output of esxtop appears normal ... "

Jürgen

[AndreTheGiant]

but the VM's uses approximately 8 - 25% of cpu resources

You have those datas from ESX or from VM Performance Monitor?

Andre

[jbauriede]

Hi Andre,

it comes from esxtop

regards

Jürgen

[AndreTheGiant]

Try to use Performance Monitor inside VM.

Just to be sure that the data are fine.

Andre