70. PREDICTABLE EXECUTION ADAPTIVITY THROUGH EMBEDDING RECONFIGURABILITY INTO STATIC MPSOC SCHEDULES

Department: Computer Science & Engineering
Faculty Advisor(s): Alex Orailoglu

Primary Student
Name: Chengmo Yang
Email: c5yang@ucsd.edu
Phone: 858-822-2505
Grad Year: 2009

Abstract
The computing engines of many current applications are powered by multicore architectures, which promise significant speedup but induce increased reliability problems as a result of ever growing integration density and chip size. While test techniques can improve shipped product granularity, they fall short of precluding the increasing occurrence of intermittent faults. Heat buildup, as it can make cores temporally unavailable during execution, furthermore degrades chip resource availability, an issue already made critical by fault occurrences. Traditional replication-based fault tolerance methods surrender the benefits of integration to a large extent, necessitating new approaches capable of delivering first rate performance through static compiler techniques, yet able to dynamically react to resource variations. To achieve this goal, in this paper we propose to statically generate a compact schedule with predictable response to various resource availability constraints. Such schedules are generated by adhering to a novel band structure, capable of spawning dynamically a regular reassignment upon resource variations. Through incorporating several soft constraints into the original scheduling heuristic, the proposed technique can furthermore exploit the inherent timing slack between dependent tasks, thus retaining the spatial and temporal locality of the original schedule. The efficacy of the proposed technique is confirmed by incorporating it into a widely adopted list scheduling heuristic, and experimentally verifying it in the context of single processor deallocations.

Related Files:

1. chengmo_CODES07.pdf

« Back to Posters or Search Results