It all depends on your workload type and on how virtualized you want to roll out the p6. E.g. if you have a lot of DB workload and want to maximize sharing of CPU and I/O, then you need to *drench* the CPUs in enough memory.
VIO ethernet also needs to have its share of memory. So if you have 2 sets of VIO servers (1 set for prod, one set for dev/test) then be sure to count at least 4GB (better 8GB, 2Gb per VIO server) extra.
I'd go for 8GB per CPU.
Power CPU's grow faster and faster at (re)incarnation, so that means they can handle more cycles and you need the memory to run all those processor cycles and to hold all the data.
If you want to drive 16 porsches at top speed at the same time, you're going to need a very wide highway to accomplish that.
And you should add up to 2GB for power hypervisor, depending on the number of LPARs. The System Planning Tool (new name for LPAR Validation Tool) can give you an estimation of the memory needed by the power hypervisor.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.