Rapid Parallel Systems Deployment: Techniques for Overnight Clustering

Automated system deployment frameworks and configuration management systems have been in wide use for a number of years. However, due to increasing pressures to maintain high availability, coupled with the price effects of commodity servers, administrators may be required to deploy large numbers of systems in shorter time frames than is normally possible with available staff. In this paper, we describe a straightforward procedure using commonly-available infrastructure to enable rapid simultaneous deployment of hundreds of machines by temporary staff. As an example of the efficacy of this approach, we present a case study in rapid systems deployment at Purdue University. On May 5th, we deployed Purdue's "Steele" cluster, installing over 500 compute nodes in a single business day.