Energy-Efficient Performance-Aware Fair Memory Access Scheduling on Multicore Platform (EEPAF)

Abstract

In current scenario, energy consumption, performance and capacity of the main memory system are key factors that affect the design of a computing system. These days, computing systems are facilitated with multiple cores. Multicore system enables simultaneous execution of multiple applications. These concurrently running applications interfere at main memory. Main memory is a major resource demanded by running threads because it stores data structures that are required for execution of an application. Main memory energy consumption and performance can be improved by reducing the number of operations required to access its memory contents and by limiting the delay to service the memory access. It can be achieved by intelligently scheduling the memory requests and it is underlying memory access scheduler that decides the scheduling of memory accesses. This paper proposes a memory access scheduling scheme, EEPAF, for reducing the energy consumption and improving the performance of main memory. EEPAF, prioritizes reads over writes, exploits row buffer hits, increases bank level parallelism, implement delayed write drain policy and ensures fairness among threads. The results quantify the main memory energy consumption for different workloads under varied core environment and demonstrate significant reduction in power consumption, energy-delay product, and execution time, while improving performance.