Found this pdf about how lpar weights & logical processors work together.
I guess there are some inaccuracies in the the number of logical processors mentioned on the link, but it gives a pretty good idea.
Joined: 06 Jun 2008 Posts: 8217 Location: Dubuque, Iowa, USA
I'm glad you found that link. The last two paragraphs are particularly worth noting:
This calculation is always applicable - even when the LPAR runs at less than 100% capacity. If an LPAR does not
use its allocation, the extra CPU cycles are re-allocated based on existing weights defined to other uncapped LPARs
requesting more CPU. However, capped LPARs cannot acquire more CPU cycles than their assigned weight, even if
those cycles are available.
With dynamic timeslicing, the LPAR weight is a guaranteed minimum, not a maximum allocation CPU resource. If all
LPARs use their allotted share, this would be the amount of processing that could be performed. If the A1-A11
LPARs had little activity, the A12 LPAR could get as much as 90% of each logical engine in its assigned time.
A few points to be VERY clear on:
1. WLM does its best when the LPAR is running at 100% utilization, but it will do what it can when the LPAR is less used.
2. Capping an LPAR impacts the overall results.
3. Dynamic timeslicing also impacts the overall results (especially when some of the LPARs are under-utilized).
4. Giving an LPAR 10% of the system does NOT mean the LPAR is going to always get 10%; sometimes it will get less and sometimes it can get more (unless capped).