Hi,
Last weekend I p2v'ed two old citrix metaframe 1.8 server running on Windows 2000 TS edition.
For some reason these two VM's consume a lot of CPU resources while the two vCPU's in the o.s. are almost idle. (Average ~10%)
I find it hard to believe that this comes from the fact that I've assigned two vCPU's to the VM's.
I've got other Windows 2000 VM's (Not TS editions) with two vCPU's and they don't consume this much resources.
One more thing: the load started to increase when production started around 7 in the morning. Before that CPU usage was idling at about 300-400 MHz.
Ideas?
WB.
if you are to look into windows task manager, can you tell if there are other services/tasks that take up the resources too?
Hi,
I've used perfmon to measure the vCPU load over a longer time span.
With the taskmanager you see less then a minute of cpu usage.
Wb.
This is also a problem when using only 1 vCPU.
Could this be related to the special HAL that was installed when the original hardware was upgraded from 1 to 2 cpu's.
I recently battled this for a few months and finally resorted to reloading new VM's from scratch as the issue was related to the P2V process and "idle loop".
We converted several Win2k/citrix PS4 physical servers and noticed that ESXI would claim the 2 vCPU's were spiked near or at 100% and the OS perfmon would show 5% usage, just like you're seeing.
The issue goes away if you reduce the VM's to 1vCPU, change the system to uniprocessor and add in the "idle loop" vmx entry, but this wasn't possible for us due to the number of users.
Some info here: http://communities.vmware.com/thread/210475 and even a thread i started some time ago http://communities.vmware.com/thread/307668.
I hope you find a better solution, but VMware support wasn't able to solve the problem and keep two vCPU's on the VM.
we've found P2V'ing citrix servers is generally a bad practice.
Hi Monderick,
Thanks for your input.
I've just changed one of our Citrix server to APCI UniProc, tomorrow I'll know the result.
For us running with a single vCPU will probably be not a problem. The original hardware had only 2 PIII based xeons @1200MHz.
Did you ever try the MPS Multiprocessor PC hal?
WB
VMware support had me try a few different HAL CPU configurations to no avail. Only changing the VM's to a single vCPU and using the uniprocessor system setting brought the ESXi CPU usage into alignment with what the OS was reporting.
Did you make any changes to the p2v process to get the different result?
I don't really like the idea of adding that "idle loop" option since it might impact the responsiveness of the server.
(not something you want on a citrix server)
We tried numerous scenarios; P2V with retaining 2 CPUs, P2V with reducing to 1 CPU leaving the HAL at multiprocessor, P2V with reducing to 1 CPU changing the HAL to uniprocessor. Even tried repair OS installs post P2V process as changing the HAL manually isn't suggested.
Nothing outside of 1vCPU, uniprocessor HAL, and vmx "idle loop" entry on the VM's would produce positive results for us.
As mentioned however, the servers were too busy not to require 2 vCPU's and we had zero success in any scenario to get the host CPU/guest vCPU performance to line up.
Changing to the APCI UniProc HAL fixed the problem for me.
I didn't need to change the "idle loop" setting.
System seems to be running just fine with only one vCPU