N.: RPJ: producing fast join results on streams through rate-based optimization

We consider the problem of “progressively ” joining rela-tions whose records are continuously retrieved from remote sources through an unstable network that may incur tem-porary failures. The objectives are to (i) start reporting the first output tuples as soon as possible (before the partici-pating relations are completely received), and (ii) produce the remaining results at a fast rate. We develop a new al-gorithm RPJ (Rate-based Progressive Join) based on solid theoretical analysis. RPJ maximizes the output rate by op-timizing its execution according to the characteristics of the join relations (e.g., data distribution, tuple arrival pattern, etc.). Extensive experiments prove that our technique de-livers results significantly faster than the previous methods. 1