GENERATING QUALIFIED PLANS FOR MULTIPLE QUERIES IN DATA STREAM SYSTEMS

Abstract

Because stream processing applications operate under strict time constraints, most previous research focuses on enhancing real-time response by decreasing a single cost function. Although such research generates an optimal or near-optimal query execution plan under the selected cost function, either in terms of memory or CPU resources, the generated plan may not actually qualify for real execution; that is, although it is optimal in minimizing the chosen memory (CPU) cost function, it may exceed the available capacity of the CPU (memory) resource. A plan is not qualified if it is optimal or near-optimal in one resource usage, whereas it is out of bound in the other. These kinds of plans are not viable in stream processing applications because one of the resource usages, either CPU or memory, will hinder the system in processing queries. This thesis proposes a technique that generates qualified global plans for multiple queries under constraints for both CPU and memory resources by scheduling MJoin and BJtree operators while sharing common operations and their results among queries.