Improving Fairness in Memory Scheduling Using a Team of Learning Automata
Aditya Kajwe, Madhu Mutyam · 2014
Conventional memory controllers deliver relatively low fairness partly because they do not learn from their past decisions. This paper proposes an intelligent memory scheduling technique for multiple memory controllers. The technique is decentralized and the controllers implicitly cooperate with each other without any exchange of information among them. Our learning technique on a 16-core simulated system gives 3.12% improvement in harmonic speedup for PARSEC workloads and 1.88% for SPEC CPU2006 workloads over Thread Cluster Memory Scheduling algorithm.