Skip to Main Content
We have developed a data stream management system that supports declarative stream queries running over high data volumes in a supercomputing environment. To enable specification of massively parallel computations our query language provides processes as query language objects. The queries call process construction functions that execute stream sub-queries assigned to a CPU. Such queries can be used to define query functions that parallelize computations. The CPU assignment is normally automatic, but can also be influenced by the user. We show how this enables performance measurements of different communication topologies in a heterogeneous hardware environment containing a Linux cluster and a BlueGene.