Reliability-Aware Core Partitioning Strategies for Multicore Architecture

Chip multiprocessors (CMPs) are fast replacing conventional single-core machines in most application domains. Core partitioning is a critical problem in the context of CMPs, which deals with how many cores to allocate to each application running simultaneously. A number of recent studies has addressed partitioning and/or scheduling related issues in CMPs where they mostly target on performance speedup or quality of service (QoS). However, there is no reliability-driven core partitioning strategy for multithreaded applications on CMPs proposed in the literature yet.

In this project, we propose and evaluate reliability-aware core partitioning schemes for multicore architectures. We perform a simulation study with various workloads consisting of multiple multithreaded applications to evaluate our proposed partitioning schemes. Our study compares the different schemes in terms of both performance and reliability. To consider both high performance and reliability objectives, a novel metric called vulnerability-delay product (VDP) is presented as part of this study. The VDP metric is a combined metric that includes both execution time for representing the performance and Thread Vulnerability Factor (TVF) for representing the vulnerability of the system to the soft errors. We also validate our partitioning schemes by executing applications and applying the strategies on real multicore architectures.