FROCM: A Fair and Low-Overhead Method in SMT Processor

Abstract

Simultaneous Multithreading (SMT)[1][2] and chip multiprocessors (CMP) processors [3] have emerged as the mainstream computing platform in major market segments, including PC, server, and embedded domains. However, prior work on fetch policies almost focuses on throughput optimization. The issue of fairness between threads in progress rates is studied rarely. But without fairness, serious problems, such as thread starvation and priority inversion can arise and render the OS scheduler ineffective. The fairness research methods always disturb the threads running Simultaneous, such as single thread sampling [4]. In this paper, we propose an approach FROCM (Fairness Recalculate Once Cache Miss) to enhance the fairness of running multithreads in SMT processor without disturbing their running states. Using FROCM, every thread’s IPC_{approximately} is re-calculated in SMT processor Once Cache Miss, IPC_{approximately} is the approximately value of IPC when the thread runs alone. Using IPC_{approximately}, the instructions’ issue priority may be changed in due course. We can hold the Fairness value (Fn) higher. Fn is fairness metric defined in this paper, when it is equal to 1, it means utterly fair. Results show that using FROCM, we can hold the most of Fn larger than 0.95, and the throughput hasn’t larger change. It needs less hardware to realize the FROCM, including 4 counters, 1 shifter and 1 adder.

This work is supported by National Natural Science Foundation of China(No. 60473079) and the Research Fund for the Doctoral Program of Higher Education (No. 20059998026).