Workload Characterization for shared resource management in multi-core systems

The multi-core industry is facing a number of challenges which need to be addressed to achieve optimal performance from the multiple cores. One of the dominant causes of performance degradation that arises in multi-core processors is due to hardware resource sharing between cores. The degradation in performance is largely dependent on the nature of the applications. To minimize this degradation, application characteristics need to be analyzed and based on this, a solution can be devised to reduce degradation. This paper aims at presenting a consolidated survey of present classification schemes and propose a new classification scheme which can be applied on-line. Also, using this scheme, a dynamic cache partitioning mechanism has been proposed. The experimentation has been conducted by constructing workloads using the SPECCPU 2006 benchmark suite and experimentation has been conducted on SIMICS (full system simulator).