DYNAMIC MULTICORE RESOURCE MANAGEMENT: A MACHINE LEARNING APPROACH A MACHINE LEARNING APPROACH TO MULTICORE RESOURCE MANAGEMENT PRODUCES SELF-OPTIMIZING ON-CHIP HARDWARE AGENTS CAPABLE OF LEARNING, PLANNING, AND CONTINUOUSLY ADAPTING TO CHANGING WORKLOAD

......Over the last five years, multicore architectures have emerged as the primary mechanism to reap the benefits of Moore’s Law in the billion-transistor era. If CMOS scaling continues to follow Moore’s Law, multicore systems could deliver as many as twice the number of cores (and cache space) with every technology generation. Unfortunately, not all multicore resources scale as easily. For example, whereas these systems ’ off-chip bandwidth requirements will likely grow proportionally to the number of cores, the number of signaling pins in con-ventional packaging technologies grows by only 10 percent with each technology gener