PEP2.0 currently supports TWO generations of Power hardware (like Power8+9, Power9+10 of Power10+11) in one pool. Suppose you currently have E950 and E1050 in one Enterprise Pool, adding an E1150 to the same pool is not possible. If you want to keep all systems in a PEP pool, you need to create a new pool, either for just the E1150's or move the E1050's and E1150's to the new pool (leaving only the E950's in the original pool). The reason is probably that newer systems have higher performance and a certain number of credits would give you same amount of CPU, but the CPU's are more powerful, so mixing POWER9 and Power11 in one pool is not currently allowed.
From a lifecycle perspective this is problematic. Customers may choose to run their systems 5 years or even longer, and new generations often occur in that time period. So it is very likely that IBM brings out Power11 while you still have Power9 and Power10 systems around. What we would like to do is introduce Power11, migrate partitions and decommission the Power9's. If the three generations would be allowed in one pool, this would be fairly easy.
In the current situation, where three generations is not allowed, this is difficult because:
- we opted for a higher base capacity for the initial (POWER9) systems, adding minimal base capacity on Power10 (only what was additionally needed)
- if we move the POWER 10 to a new pool (with Power11), we need to buy the Power11 with high base capacity, because we lack the Power9 base (remains in the original pool, base largely unused)
- if we just put the Power11 in a new pool, we could start with modest base, but that would need to change one we start migrating workloads from Power9 or 10 there
It becomes jiggling of partitions around, taking care to do proper calculations of base capacities, expected usage exceeding the base and having enough credits to pay for over-base usage. We need to do this for every power9 that is decommissioned and its load distributed over existing Power10 or new Power11.
The HMC versions support THREE generations of Power, V11 of the HMC code will support Power9, 10 and 11 (but no more Power8), something we really need PEP2.0 to do as well.
For the problem that 1 credit will buy you a certain amount of CPU (or a certain amount of CPU costs 1 credit to be exact) and it is unfair to count Power9 and Power11 the same, there are three solutions:
- Make the amount of CPU (you get for 1 credit) higher for P9 and P10 (so P11=one, the others scale according to their relative per-cpu performance);
- Make the amount of CPU (you get for 1 credit) lower for P11 (where the P9=1, and P10=1 when no more P9 present in the pool, scaling according to their relative per-cpu performance)
- Keep it simple and keep the amount of CPU (you get for 1 credit) identical. Meaning a P9 will cost you the same as a P11 (in terms of credits), which could help moving workload to P11 and getting rid of the P9 sooner.
So the bottom line is: please make PEP2.0 behave as the HMC does, supporting three generations of POWER servers in one pool.
