Click here for full text:
The Performance of Self-Scheduled Concurrent Systems
Tofts, Chris
HPL-2001-121
Keyword(s): scheduling; Markov Chain; concurrency; embedded systems; transients
Abstract: The solution of complex real time tasks is often achieved by pipe-lining. The task is broken up into several `smaller' blocks which may share both processing and memory resources. When these block execution times are variable, and consequently difficult to schedule, a standard solution is to execute computational elements concurrently and allow them to `find' a good schedule as a consequence of their interactions. In this paper we examine how long this search may take, and demonstrate that even in relatively ideal circumstances this time can be comparable with the total task duration. We then demonstrate a simple technique that reduces the effects of this problem without requiring the imposition of a global schedule.
28 Pages
Back to Index
| 
printable version
