If i may add,
you'll also see high Time waited for prefetch (ms)" when you're too
agressively prefetching.
e.g. prefetch queue(s) are full, you're asking too many io's per sec (there
the wait).
Also if you prefetch more than the bp size you use, you compete for bp
space.
You should see something like stolen pages or forced writes, cleaners
triggered and waits, etc.
If you want to investigate more, i suggest you read a bit on DB2BPVARS.
http://publib.boulder.ibm.com/infoce...n/r0005665.htm
A good and simple way to measure the io performance is the old ms/io.
You start low and increase and see what it does on a tablescan or your
workload.
There would be a relation with the number of io's and the size you're
asking. (containers setup, extent and prefetch values, etc.)
In the end, every environment is different. (io capacity of adapters used,
etc.)
A general rule of thumb is NUM_IOSERVERS = nb cpu + a few extra.