An Energy-Aware Workload Dispatching Simulator for Heterogeneous Clusters

This paper presents an energy-aware workload dispatching simulator for heterogeneous clusters. Most clusters in a data center are composed of different kind of machines. Among these machines, the front-end nodes distribute incoming requests to the back-end workers. The main concern in such system traditionally focuses on computation performance, but energy consumption has emerged as an equally important issue recently. This work designs a workload dispatching simulator that is capable of identifying the energy usage of a heterogeneous cluster. The main target of the proposed simulator is to find out how a workload dispatching algorithm affects the energy consumption. The simulator consists of a configurable traffic generator, a workload dispatcher, and a set of server nodes. The traffic generator randomly produces workloads that are constrained by the average job number, and the average job size. The workload dispatcher assigns jobs according to some popular workload dispatching algorithms. Each server node can report its ID, idle power consumption, maximum power consumption, and computing capability. A set of records collected from a real heterogeneous cluster is examined using this simulator. The goal of this work is to develop a simulator that assists data center managers to estimate the availability, service quality, and energy consumption of their heterogeneous computation resources.