A Programming Model and Language Implementation for Error-Tolerant Networks of Computation

Abstract

Many embedded applications involve processing data from noisy analog signals and displaying information for human observation. In such systems, trade-offs often exist between performance, energy usage, and the accuracy of data processing. To achieve the combination of both low idle-power consumption and high peak performance often required in embedded systems, there is an increasing trend towards the use of multiple processing elements instead of a single high performance processor. Combined with the presence of correctness-performance trade-offs, this trend provides interesting new programming language challenges.

This paper introduces a programming model, its runtime system, and a language implementation, targeted at systems containing multiple resource-constrained processors which may be able to trade off performance or power consumption, for correctness and timeliness of computation. The programming model makes it straightforward to achieve partitioning of program implementations across the code storage memories of multiple resource-constrained processors, and introduces the idea of program-level constraints that permit compile- and run-time trade-offs between performance and correctness to be exercised. The ideas are presented in the context of two different hardware platforms we developed that benefit from the facilities of the programming model.

Cite as

P. Stanley-Marbell. “A Programming Model and Language Implementation for Error-Tolerant Networks of Computation”. Technische Universiteit Eindhoven, TU/e ES Tech Report, Number ESR-2008-03, 2008.

BibTeX

@article{stanley2008programming,
title={A programming model and language implementation for error-tolerant networks of computation},
author={Stanley-Marbell, Phillip},
journal={ES reports},
volume={2008},
year={2008},
publisher={Technische Universiteit Eindhoven}
}