A Novel System-Level On-Chip Resource Allocation Method for Manycore Architectures

Abstract

Manycore architectures are expected to be the dominant trend in future general-purpose computing systems. With the number of on-chip processor cores rising to the hundreds, the problem of resource allocation cannot be addressed with traditional methods employed by off-chip multiprocessor architectures. We propose the use of system-level bidding-based algorithms as an efficient and real-time on-chip mechanism for resource allocation in manycore architectures. We simulate and evaluate the proposed bidding-based algorithms in a hierarchical, on-chip network connected manycore platform. Experimental results indicate performance improvement between 15-25%, when compared to an on-chip round robin allocation, while achieving a balanced workload distribution. The obtained results encourage further investigation of the applicability of such system-level algorithms for addressing additional important problems in manycore architectures.