Machine Learning for Multicore Power Management

Proposed design for learning power saving opportunities.

There is an increasing complexity in applications: intensive computation, heavy network transmission, always on display, streaming apps, varying usage patterns especially with increasing complexity of architecture: multi-core, many-cores, FPGA, shared memory. Even with the power management methods like DVFS in place, if an application causes the processor to perform poorly, that will have a negative effect on power utilization.

New power saving opportunities are present in the form of application or workload(multiple application) characteristics’ that drive power usage at any point in time and which has a strong correlation with user behaviour and system state at that point in time. Our work involves coming up with interesting models that learns from experience to better control the power state-space depending on application workload, user behaviour and system state.