Evaluation of Join Strategies for Distributed Mediation

Three join algorithms are evaluated in an environment with distributed main-memory based mediators and data sources. A streamed ship-out join ships bulks of tuples to a mediator near a data source, followed by post-processing in the client. An extended streamed semi-join in addition builds a main-memory hash index in the client mediator. A ship-in algorithm materializes and joins the data in the client mediator. The first two algorithms are suitable for sources that require parameters to execute a query, as web search engines and computational software, and the last is suitable otherwise. We compare the execution times for obtaining all and the first N tuples, and analyze the percentage time spent in subsystems, varying the network communication speed, bulk size, and data duplicates. The join algorithm leads to orders of magnitude performance difference in different mediation environments.