CPU Ready ?

[rowl]

I am having some performance issues with some of my Spectrum Protect servers which are running in VMs. So far the only thing that I see that appears to be amiss is a performance metric called CPU Ready in the vSphere client. This is a measure of time when the VM is ready but cannot get scheduled to run on a physical CPU. I am seeing hours of time where this is in the 4 to 10 second range. What's odd is that the physical host has 44 cores, the VM 20 vCPUs and there is no other VM on the host.

 

[moon-buddy]

What kind of performance issues are you getting?

I have been a very vocal proponent on this Forum about not running TSM in a VM machine EVEN IF it is the only one there. The reason is that VMs rely on hypervisors for managing resources. TSM is I/O intensive for most of its Operations. Hypervisors adds a complexity layer and TSM cannot fully take adavantage of the unobstructed view of the I/O system.

Thus, I don't use VMs for TSM instances, and this has been a battle that I have won time and time again.

 

[rowl]

I wish I had that option to not run on a VM, but that's a non starter here. You WILL use a VM or you know where the door is more or less. I have 24 SP servers running VMs and for the most part I am not having problems with them.

This problem was identified as an issue with a VM setting for CPU hot add that was not enabled. Once this was updated and the system restarted these CPU busy issues ended. This setting was enabled on all other SP VMs, so human error I guess.

Now I am having other issues, and the VM layers of abstraction certainly make troubleshooting this difficult. Occasionally one of the virtual disks will have latency from 50 to 500ms for hours at a time. Working with the VM and SAN teams to try aid figure out what is going on there. The disks are either used for the DB2 logs or one of the DB file systems.

 

[moon-buddy]

I wish I could offer more help but I have gone this route before and proved to the powers that be that TSM is not meant to be run in a VM especially VM Ware or Hyper V.

Even when I tried TSM on AIX on top of a hypervisor, the performance was terrible. TSM on AIX as a LPAR that has predefined resources works but still not to my liking.

One thing to note, remove CPU HOT Add and stick to a defined CPU and resources count. CPU HOT Add allows CPU sharing which can drastically punish VMs that are CPU and I/O intensive when these VMs are at their lull moments.

 

[marclant]

Source: https://www-01.ibm.com/support/docview.wss?uid=swg21239546