Auto-Pipe: Streaming Applications on Architecturally Diverse Systems

Abstract

Architecturally diverse systems can improve streaming application performance by orders of magnitude, albeit with enormous programmer effort. To simplify the programming of such systems, we have constructed the Auto-Pipe application development environment, which supports the flexible mapping of application components onto computational resources and the automatic delivery of data between these computational resources. In addition to the computational finance application, Auto-Pipe has been used to implement applications ranging from cryptography to astrophysics. Auto-Pipe currently supports applications deployed on chip multiprocessors and FPGAs, and we are expanding its scope to include graphics engines.