IaaS Procurement by Simulated Annealing

Considering the problem of resource allocation for potentially complex and diverse streaming (e.g., query processing) or long-running iterative (e.g., deep learning) workloads in the public cloud, we argue that a framework based on simulated annealing is suitable for navigating performance/cost trade-offs when selecting from among heterogeneous service offerings. Annealing is particularly useful when the complex workload and heterogeneous service offerings may vary over time. Based on a macroscopic objective that combines both performance and cost terms, annealing facilitates light-weight and coherent policies of exploration and exploitation when considering the service suite offered by a cloud provider. In this paper, we first give some background on simulated annealing and then demonstrate through experiments the usefulness of a particular resource management framework based on it: selecting the types and numbers of virtual machines for a particular job stream.