KURT-Linux support for synchronous fine-grain distributed computations

Synchronous fine-grained distribution is an important aspect of an interesting range of applications. Most conventional platforms for supporting computations of this type constrain their performance because of the lack of scheduling integration and coarse timing granularity imposed by the underlying operating system. This paper discusses the use of a real-time and embedded system approach to improve the performance of such distributed computations by explicitly scheduling their components, by providing fine-grain temporal resolution, and by producing and consuming messages among the distributed computation components in a just-in-time manner. In the motivating application for this work, KU real-time (KURT) modifications to the Linux operating system were used to control synchronized distributed components of ATM network simulations. Metrics for evaluating the performance of generic synchronous distributed computations are discussed and results are presented for the KURT-Linux based system.