Design And Analysis Of Location Cache In A Network On Chip Based Multiprocessor System

In recent years, the direction of research to improve the performance of computing systems is focused toward chip multiprocessor (CMP) designs with multiple cores and shared caches integrated on a single chip. To meet the increased demand for data, large on-chip caches are being embedded on the chip, shared between the multiple cores. The traditional bus-based interconnect architectures are non-scalable for large caches and cannot support the higher cache demand from multiple cores, which motivates the design of a network-on-chip (NoC) interconnect structure for shared non-uniform cache architecture (NUCA). The concept of NUCA caches proposes the division of the cache into multiple banks connected by a switched network that can support the simultaneous transport of multiple packets. The larger on-chip cache designs also result in higher power consumption which is a serious concern as fabrication scales down to the nano-technologies. This research focuses on the implementation of the location cache design in a NoC-based NUCA system with multiple cores, in combination with low-leakage L2 cache based on the gated-ground technique. This system architecture helps to reduce the power of L2 cache along with the performance benefit of the on-chip network. The CMP cache system is implemented on a NoC-NUCA framework with a write-through coherency protocol. The features of CACTI and GEMS are extended to support a complete power and performance estimation of the system. A full-system simulation is performed on scientific and multimedia workloads to characterize the NoC-based system. An analysis of the power and performance of the proposed system is presented in comparison with the traditional cache structure in different configurations. The simulation results show that the NoC-based system with the location cache results in significantly saving the energy of the cache system over the traditional bus-based system in any configuration and also the NoC-based system without a location cache. The system also provides better performance compared to a bus-based system, emphasizing the need to shift to a network-based cache interconnect design which can scale to a large number of cores.