Stream Based Supercomputing - The Auto-Pipe System

Stream Based Supercomputing is an interdisciplinary research group established at Washington University in St. Louis. We concentrate on large-scale scientific applications, particularly those that are characterized by high-throughput stream processing requirements. Our membership includes computer engineers and computer scientists, as well as expert scientists in application areas such as astrophysics, networking, and bioinformatics.

Our current efforts concentrate on improving the performance of scientific applications in astrophysics and bioinformatics, especially those that operate on large streams of data from arrays of disks or sensors. We achieve these performance improvements through

• exploitation of parallelism and pipelining in streaming algorithms across multiple computation nodes,
• taking advantage of the relative strengths of certain special-purpose computing technologies (e.g. FPGAs and Network Processors) in the appropriate stages of computation,
• partitioning the application into sets of tasks and efficiently mapping the tasks onto computation nodes, and
• developing tools to analyze applications and aid the developer in performing the above operations.

The work done by the Stream Based Supercomputing group is primarily supported by the National Science Foundation through grant CCF-0427794.