Parallel Query Scheduling and Optimization with Time and Space-Shared Resources
Scheduling query execution plans is a particularly com- plex problem in hierarchical parallel systems, where each site consists of a collection of local time-shared (e.g., CPU(s) or disk(s)) and space-shared (e.g., memory) re- sources and communicates with remote sites by message- passing. We develop a general approach to the prob- lem, capturing the full complexity of scheduling distribut ed multi-dimensional resource units for all kinds of paral- lelism within and across queries and operators. We present heuristic algorithms for various forms of the problem, some of which are provably near-optimal. Preliminary experi- mental results confirm the effectiveness of our approach.