Cloudburst - Simulating Workload for IaaS Clouds

Abstract

In this work we implemented Cloudburst to generate realistic workload in Infrastructure as a Service cloud testbeds. Our goal was to minimize the memory footprint of such workload generators by leveraging alternative programming paradigms for highly concurrent applications. In contrast to many existing we leverage the concurrency model of the Go programming language instead of threads. Initial benchmarks with Cloudburst and Rain suggest that Cloudburst consumes significantly less memory at the cost of a higher CPU footprint. In our experimental testbed memory is a more critical resource. Leveraging Cloudburst allows us to run larger benchmarks with the same hardware.