An optimistic approach to computational steering

Abstract

Computational Steering is the online, interactive allocation of resources and adjustment of application parameters. Few computational steering systems support the coordinated steering of multiple processes. Of those that do provide such support, our system is unique in its optimistic approach; other systems take a conservative approach. Because this requires global synchronization considerable perturbation can arise. We focus on optimistic steering, which does not require global synchronization before a steering event may take place. To achieve this requires not only the ability to determine the consistency of steering transactions but also the ability to correct any inconsistencies that may occur. To address these issues, we have developed algorithms for consistency detection and a steering system that has the ability to correct inconsistencies through computational rollback and re-execution. Presented in this thesis are both the details of our steering system and a performance analysis of that system.