A Distributed Application-Level IT System Workload Generator

Abstract

Developing a capacity to test distributed systems hinges on being able to generate the workloads that these systems are to process. Appropriate tools must not only generate these workloads in real-time, but must also be able to sweep through a range of possible workload characteristics to support sensitivity and robustness analyses. Currently, the majority of prior work in this area, including Harpoon, ns-2, OpNet, and tcp replay, has focused on the reproduction of workload traces at the network-level. However, for many distributed systems, reproducing application-level workload characteristics is more informative from a testing perspective. This work details such an application-level workload generation tool. The tool itself is distributed and, hence, easily scales to using multiple machines to re-create complex multi-homed workloads. Furthermore, the tool supports the standard abilities to produce both statistically-described workloads, as well as reinstantiating previously-captured workload traces.