Edit: Although we are working with POWER9 instead of POWER6, these questions and discussions are still ongoing today.
Originally posted August 19, 2008 on AIXchange
A vendor wants to charge you a “per CPU license” fee. What number are vendors looking for when they want to count CPUs in your machine to calculate what you owe? Are they looking for the number of physical sockets on your machine, or maybe the number of processor cores? Are they looking for the total number of processors installed, or do they only count the number of CPUs that are actually activated on your machine?
With the advent of Micro-partitioning, there’s another thing to consider. Are they looking at your HMC to learn my minimums and maximums for each LPAR? Are they looking at the number of virtual processors that your operating system sees? Are they looking for the “number” of processors that you consume when borrowing CPU cycles from other, less busy LPARs during busy times?
More vendors are realizing that we’re less likely to dedicate a whole CPU to an LPAR. Depending on the workload, dedicating a CPU to a “less busy” LPAR when another LPAR could make use of those “idle” cycles can be a waste of resources.
What is a processor? When your HMC reports that 4 CPUs are available on your machine, what does that mean? Do you have four actual chips in your machine?
On my JS22 for instance, it tells me I have four processors. There are two chips on the blade; each is dual core. Is my vendor interested in the number of physical sockets on the blade? Is it interested in my processor class? My POWER6 system will do more work than my POWER5 system did, so I can configure fewer CPUs for my LPAR to do the same amount of work. Will my vendor then charge me less?
As we become more virtualized, this topic will continue to be revisited, but it’s another thing to think about as we justify upgrading our hardware. We may well be able to do more with less, and lower our software bills in the process.