John Hammond wrote:Hi another question on the unweeded garden that is TM1 licencing.
Is there any truth in rumour IBM ignores processor limit assigned to VMs when calculating PVUs and goes for the physical machine on which the VM resides.
Ie have max of 4 cores dedicated to our VM which equates to 4 * 70 = 280 PVUs, but machine has 24 cores so get hit for 24 * 70 = 1680 pvus!
Yes. And er, No.
The Yes part is actually aimed to be beneficial, but doesn't apply in your case. The theory goes that even though your host has 24 cores, it is possible for you to be hosting, say, 8 x 4 core VMs on that host with the virtualisation technology dynamically handling the core overlap. In this case licensing your VMs individually would require licensing 32 cores. As this is greater than the number of cores in your host, you only need to license the host's 24 cores. Which makes sense (not that I can believe I am saying that about IBM licensing) as even if your VMs think that they have 32 cores in total, the actual number of instructions that the cores can process is limited by the host's 24 cores.
Given that you say you have 24 cores on the host, I am guessing that you have 4 sockets with 6 cores on each. The only architectures (I'm assuming you're on Intel) in the
PVU Table that provide hexacore processors are Xeon Nehalem EX and Xeon Sandy Bridge EP, and for both of these a machine with 4 sockets rates at 100 PVUs per core, not 70. So your host is actually 2400 PVUs.
The grey area comes in around your requirement to use the
ILMT to monitor CPU assignments to the VM. Assuming you are not excluded from the need to use this tool via exception 2 on that link, my reading of exception 3 would suggest that as your host requires more than 1000 PVU (the 'total physical capacity of your server') then you should be using it.
Having said all that, my understanding would be that as your VM only has 4 cores, and as this does not require the host to use more than 2 sockets, then yes your license requirement will be 4 x 70 = 280 PVUs. Whether you think you need to use the ILMT to prove that you're keeping to the 4 CPUs is up to you.
All the usual caveats apply, and I'll be interested to see how many other interpretations of the same documents there are out there...