A Case for Merge Joins in Mediator Systems

There has been considerable work on determining efficient joins for mediator systems. Mediator sys-tems have to dynamically integrate the results of multiple sources in a timely and efficient fashion. Thus, a focus has been on the ability to produce initial results early and to handle network latency and blocking. The issue of limited client capa-bilities and its impact on join algorithm selection has not been examined. A mediator system may be a large database system in itself or as small as a PDA, cellphone, or embedded Java applet on a web page. It is these smaller, dynamic mediator systems that hold promise of rapid integration of web resources. In this work, we examine several of the existing join methods under the context of supporting thin client mediators which have lim-ited memory, CPU, or disk space. Somewhat sur-prisingly, the basic sort-merge join applies excep-tionally well to this environment with its paral-lel exploitation of server resources and minimal client resource constraints.