An Approach to Discrete Parameter Design Space Exploration of Multi-core Systems Using a Novel Simulation Based Interpolation Technique

Abstract

We propose a new approach to simulation-based design optimization of multi core systems, over a large number of discrete parameters. In this approach, we embed the discrete parameter space into an extended continuous space and apply continuous space optimization techniques over the embedding to search for optimal designs. Such continuous space techniques often scale well with the number of parameters. The embedding is performed using a novel simulation-based ergodic interpolation technique, which, unlike spatial interpolation methods, can produce the interpolated value within a single simulation run irrespective of the number of parameters. In a characterization study, we find that the interpolated performance curves are continuous, piecewise smooth and have low statistical error. We use the ergodic interpolation-based approach to solve a multi-core design optimization problem with 31 design parameters. Our results indicate that continuous space optimization using ergodic interpolation-based embedding can be a viable approach for large multi-core design optimization problems.