An interesting albeit rare scenario that we have seen occur when using CPU limits in a shared LPAR which resulted in a downward “death spiral” with the hypervisor continuosly reducing the amount of CPU available to DB2 and severely affecting DB2 performance.
The Workload Manager dispatcher technology introduced in DB2 10.1 offers a powerful and simple interface for controlling CPU. The two controls used to provide direction to the WLM dispatcher are CPU limits and CPU shares. Ever wonder how to choose between them?
We are collecting statistics every two minutes but looking at the CoordActLifetime histogram data collected in one of these 2 minute windows, we observed that in the upper bins, there are activities with lifetimes of 1715085 milliseconds (@28 minutes). How is this possible?