Topics of the chapters

We explain where memory-hierarchy aware techniques are usefully applied in the field of parallel programming. The discussion includes distributed, cluster and SMP architectures, that may be nested hierarchically. Actually, several resources of a parallel system can be seen as different layers of a memory hierarchy, like shared and distributed memory, or local and remote disks managed by conventional / parallel file systems.

The emphasis is on the software aspect of employing HM techniques on hierarchically structured parallel machines (other chapters cover the issues of CPU-level caching and prefetching). Relevant topics are

• issues at the different levels that suggest or may prevent the use of HM techniques

• what software tools are available at the different levels

• some external memory models (which we assume are explained in other chapters) and their connection to bridging models for parallel computation

• what HM solutions are employed in the various tools.