Efficient Resource Utilization in Shared-Everything Environments

Efficient resource usage is a key to achieve better performance in parallel database systems. Up to now, most research has focussed on balancing the load on several resources of the same type, i.e. balancing either CPU load or I/O load. In this paper, we present floating probe, a strategy for parallel evaluation of pipelining segments in a shared-everything environment that provides dynamic load balancing between CPU- and I/O-resources. The key idea of floating probe is to overlap---as much as possible with respect to data dependencies---I/O-bound build phase and CPU-bound probe phase of pipelining segments to improve resource utilization. Simulation results show, that floating probe achieves shorter execution times while consuming less memory than conventional pipelining strategies.