Thinking Big in a Small World - Efficient Query Execution on Small-scale SMPs

Many techniques developed for parallel database systems were focused on large-scale, often prototypical, hardware platforms. Therefore, most results cannot easily be transfered to widely available workstation clusters such as multiprocessor workstations. In this paper we address exploitation of pipelining parallelism in query processing on small multiprocessor environments. We present DTE/R, a strategy for executing pipelining segments of arbitrary length by replicating the segment's operator. Therefore, DTE/R avoids static processor-to-operator assignment of conventional processing techniques. Consequently, DTE/R achieves automatic load-balancing and skew-handling. Furthermore, DTE/R outperforms conventional pipelining execution techniques substantially.