General scheduling domains take 2
Hi,
http://www.kerneltrap.org/~npiggin/w24/
I have progressed a bit further, and the data structure is slightly changed.
I have done code to build the scheduling description for SMT P4s. It can be enabled with a new config option although don't try this with NUMA yet (easy to fix). Note this is quite unlikely to actually do any good on a pentium 4 because I haven't been able to test it!
The generic scheduling description works nicely on the 16 way NUMAQ at OSDL and gives results very close to those I was getting before, so I think I have done a good job at recreating that scheduling policy, and the scheduling domains doesn't appear to have too much overhead.
The data structure is a bit difficult to understand. I have the start of a paper here http://www.kerneltrap.org/~npiggin/paper.ps.gz but its only really worth looking at for the pictures at the end for now.
Questions, comments welcome.
Nick