Periodic pause on Guests

[Russell.Wood]

Hi all.

I have HPVM 3.00 and 3.50 installed on two rx6600 with the guests running HP-UX 11.31. These are recently fresh installs.

When connected to a Guest via the network or console, periodically the CLI freezes for a few seconds - whatever is being typed does not display and then after a few seconds all of what was typed appears.

Has anyone experienced this before?

[Russell.Wood]

I just ran two ping tests from another server to the guest and also from the host to the guest. The freeze is every minute. The ping goes from 0ms to ~800ms.

[Eric SAUBIGNAC]

Bonjour Russell,

I suppose that you have checked health of physical card on which the virtual switch is configured ?

What about memory in the guest ? Under heavy guest memory pressure I have sometimes seen that the guest becomes frozen. In this case there is frequently no way to properly halt the VM. The whole host must reboot ...

What about free memory on the vm host ?

What about CPU utilization in the guest (sar -u --> %sys) ? I have seen that a guest can sometimes have unexplained usage of cpu in system mode. We thought, and I still think, that the source of this problem was the virtual IO stack.

The first thing you should do is to check memory on guest an host.

Second point is to use AVIO, both for virtual disks and virtual lans. You will find packages to install on host and on guest at http://h20392.www2.hp.com/portal/swdepot/displayProductInfo.do?productNumber=AVIO . To use AVIO, HPVM must have at least version 3.5

The last, but not the least : you should upgrade to HPVM 4.0 + AVIO. It works on 11iv3 hosts and HPVM will take advantage of the new 11iv3 IO stack.

I have plan to migrate all of my clients to this new version of HPVM. I think that HPVM under 4.0 version is not a mature product. So you can face some strange behaviors without being able to explain why. I still have no advice on HVPM 4.0 but I do hope will have a more stable virtualization stack

Regards

Eric

[Russell.Wood]

Hi Eric,

Both the Host and Guests are under no load at all (CPU/Memory/Swap/IO is all low).

Because this is happening to two physically different rx6600s then I'm assuming it's not a hardware fault. Both Hosts were installed fresh but the Guests installed from an image.

[Duncan Edmonstone]

Russell,

I assume you have confirmed that this is not an actual networking problem by pinging the VM host from another system?

HTH

Duncan

[Russell.Wood]

My second post:
"I just ran two ping tests from another server to the guest and also from the host to the guest."

[Eric SAUBIGNAC]

... so you miss ping between an other server and the VM host. Every ping you have done involves the guest

[Eric SAUBIGNAC]

"Because this is happening to two physically different rx6600s" --> U right, probably no hardware problem

You should try to install AVIO on the 3.5 HPVM host and check all HPVM necessary patches, host and guest

Eric

[Russell.Wood]

Turned out to be a scheduling issue on the Host. The `cimprovagt' process was spiking (using top) every minute. When I terminated the process it stopped the pause (although other processes occasionally did the same thing which caused the same pause).

I reniced the hpvmapp process so it has a high priority, restarted cim_server and all is good.

[Russell.Wood]

Also, I was running AVIO on one of the servers and the problem was still evident.

[Russell.Wood]

Closed.