Modern server machines often have sufficient main memory to keep the hot set of a database–if not the entire data set–in memory at all times. At the same time, the access characteristics of today's memory technologies are by far not "random access", in spite of the term "RAM" still used for the computer's built-in memory. Caches, bus systems, interconnect topologies, and parallelism all interact with a program's access pattern.
Together with researchers from ETH Zürich and U Waterloo, we are developing and evaluating strategies to implement database algorithms for such scenarios: massively parallel in-memory processing. We currently hold the world record in join processing in such environments.
Project collaborators: