An NBTI-aware Task Parallelism Scheme for Improving Lifespan of Multi-core Systems

Abstract

Task parallelism schemes with a multi-core system are widely used through various applications nowadays to obtain better throughput. Most of previous works will try to find maximum degree of parallelism (DOP) to achieve best performance. However, aging effects have become an unavoidable threat which may degrade system performance and even cause functional failure. We observe that if a task is always executed with maximum DOP, all cores may continuously suffer from the NBTI effect and wear out soon. On the other hand, if we can decrease DOP without causing task deadline violation, some cores may have opportunities to be power-gated and recover from the NBTI effect. In this paper, we propose a novel NBTI-aware task parallelism framework for multi-core systems to realize such an idea. Our framework first prioritizes ready tasks based on its criticality and then decide best DOP of each task without causing timing violation. After that, a task-to-core mapping algorithm which considers both IR-drop and core utilization is proposed. Finally, a recover decision algorithm is used to find the suitable recover mode of each core. Experimental results show that the proposed NBTI-aware task parallelism framework can successfully extend system lifetime to 3.7 times compared with MAX DOP method.