VMware Cloud Community
HelgeL
Enthusiast
Enthusiast

ESX CPU has 70-80% load. VM has 10%

The ESX server registers the CPU of the VM to run at 70 to 80% load, but when i log into the VM the task manager is only showing 0 - 10% CPU use. It has enough RAM, and there is no balooing going on. Why is the ESX server using so much CPU when the VM clearly is not using them?

Tags (3)
0 Kudos
14 Replies
weinstein5
Immortal
Immortal

Do you have other VMs running?

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful
0 Kudos
HelgeL
Enthusiast
Enthusiast

Yes. Right now there are 7 running VMs on the ESX. None of the others are having these difficulties. The ESX has 4 cores with 2,66GHz. Total use of CPU on the ESX is 4576MHz, so that shouldn't be the problem.

0 Kudos
BenConrad
Expert
Expert

if you look at esxtop does it show that the 70-80% load?

Do you have any uni-cpu VMs with mulit-process HALs?

Anders_Gregerse
Hot Shot
Hot Shot

One reason might be that virtualization isn't always a 1-to-1 comparision. Some calls made by a vm may translate into a high physical cpu load (depending on application and the physical cpu). Context switching is one of those things that doesn't virtualize well on all cpus. Take a look at presentation TA68 rom vmworld 2007, it will explain it in more detail.

What are the vm running?

0 Kudos
HelgeL
Enthusiast
Enthusiast

I'm not quite sure how to enterpret the output from esxtop. It shows:

% USED %SYS %OVRLP %RUN %WAIT

Idle 237.70 0.00 0.00 0.12 0.00

vmls_pacs 149.05 0.01 0.15 149.63 669.00

There are of course several other VMs on the esx. They are all showing %WAIT numbers from 300 and up to 1300 (the helper process)

The vmls_pacs is a converted machine. There is also another imported macine that is showing the sam symptoms. VMs that are created in the ESX from the start does doesn't have any performance issues.

The ls_pacs VM had 4 vCPUs. I removed two of them, and booted the VM. After one bluescreen and another reboot it shows loads of between 3 and 8% in task manager in the VM. The ESX is now using 40% of the VM's available cycles on the VM

Message was edited by: HelgeL

0 Kudos
HelgeL
Enthusiast
Enthusiast

It is running an SQL server for the Xray end MR machines at the

hospital. It is just a test convertion, the physical machine is still

doing the job, and has a load between 0 and 5 %.

0 Kudos
HelgeL
Enthusiast
Enthusiast

The VM is a win 2000 SP4. It is running the SQL service for the hospitals Xray and MR machines. The physical machine is still on and doing the job. The converted one is just for testing purposes. It has no connection to the SAN or production network.

0 Kudos
HelgeL
Enthusiast
Enthusiast

The WM is a Win 2000 SP4 running the SQL service for the hospitals Xray and MR machines. This is just a test convertionand the physical one is still online. The VM has no connection the SAN or production network.

0 Kudos
HelgeL
Enthusiast
Enthusiast

The WM is a Win 2000 SP4 running the SQL service for the hospitals Xray and MR machines. This is just a test convertionand the physical one is still online. The VM has no connection the SAN or production network

0 Kudos
Anders_Gregerse
Hot Shot
Hot Shot

Since it's a converted server. If it has only 1 vCPU please verify that it is using a uni processor hal and if it has more than 1 vCPU that it is using a multi processer hal. If it has 1 vCPU and a multi processor hal, it explains your problems and you need to replace the hal with a uniprocesser version (there are numorous articles about that).

HelgeL
Enthusiast
Enthusiast

One of the converted servers had a single CPU, the other had 4 cpu's. The single CPU has a uni CPU hal. The 4 CPU also has a uni CPU hal. I will try to change the hal of the single vCPU VM. Will changing the hal require a reboot of the VM?

The aledged CPU use of the 4CPU VM (now downgraded to a 2 vCPU VM) is down to 20% CPU usage on average and 47% max when it i not doing anything. In task manager it is still using below 10% CPU

0 Kudos
BenConrad
Expert
Expert

points? Smiley Wink

0 Kudos
HelgeL
Enthusiast
Enthusiast

i cloned the VM with single vCPU and multi hal to measure perfomance between the 1. and 2. When the clone booted up it uses almost no CPU on the ESX. When I changed the hal to uni (I used this hal : Advanced Configuration and Power Interface (ACPI) PC ). There was no noticable difference before and after. Why are the two now identical servers creating so different CPU load? (The only difference is that the 2. VM isn't connected to any switch.)

How identical are the clone? Will the SID, MAC etc be identical? I am thinking of just renaming the 2nd VM and use it as the production server.

0 Kudos
weinstein5
Immortal
Immortal

If you have Sysprep extracted to the approriate spot in VC directory structure and ran the guest customization than the SD, IP, Machine name all should be different - if you did not run the guest cutomization than all that should be identical - the MAC address should always be unique -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful
0 Kudos