I would like to give a short talk about some systems issues that are often overlooked in
discussions about using multi-core processors.  

When thinking about desktop workloads machines vary massively in the resources
devoted to a given application, either because of hardware differences, or because
many applications and services are running concurrently on the same machine.  

The challenge is not just designing programming models that enable applications to run
effectively across different levels of hardware parallelism, but also arbitrating the
conflicting requirements of concurrent applications -- e.g. I do not want my spare CPU
time used to run a disk indexing workload when the disk is already saturated.  I would
rather that some cores be left idle if that processor's memory bus is already saturated. 
If two non-time-critical background tasks can each fill all of memory then let us run them
in series rather than parallel.

What can we do to deal with these kinds of resource management problems?