Special session: memory hierarchies

Applications that process huge data sets have to take into account a memory hierarchy of caches, local memory, non-uniform shared or distributed memory and disks. This becomes more and more important because there is an widening latency gap between processors and main memory or even disks. Furthermore, new hierarchy levels like third level cache, NUMA, or clusters of SMPs are becoming wide spread. The papers in this special session represent an interesting mix of advanced topics in this area.