Poster Abstract: Automatic Deployment Right-Sizing Through Hyperparameter Optimization

Abstract

Internet of Things (IoT) and Edge deployments are diverse, complex, and highly constrained. These properties make correctness difficult or impossible to verify a priori. We present early work on an automatic deployment right-sizing tool for edge and IoT deployments. Our tool uses the PROWESS testbed to accurately emulate candidate deployment form-factors, and optimizes deployment parameters to minimize costs. We show that our early work finds optimal deployment configurations 6.3X faster than Bayesian optimization, a state of the art hyperparameter optimization technique.