University of Houston
Department of Computer Science
In partial fulfillment of the Requirements for the Degree of
Master of Science


Deepti Vyas
will defend her thesis

Global Computing on Compute Clusters

Abstract

Global computing based on utilizing unused cycles on public computers, pioneered by SETI@home project is being increasingly employed to solve important real life problems. An important source of unused compute cycles are compute clusters consisting of 10s to 1000s of nodes. These clusters are particularly attractive for global computing as they offer a homogeneous, well maintained pool of processors that is easier to exploit than disparate personal computers. Most compute clusters execute threads from one parallel application at a time on one processor. When a parallel application is executing on a group of processors, each processor is typically not completely busy.

Our goal was to experimentally determine the scale of free cpu cycles available on a typical compute cluster in production use and to see how we can utilize these cycles for a guest application without impacting the main applications. To achieve this goal, we collected performance data from the High Performance Computing cluster at UH. This data shows that even when the nodes are busy, their average cpu utilization is typically between 65-85%. Thus, these nodes have a lot of idle cpu cycles, but available at a very fine grain - in the order of milliseconds. We propose to use these idle cycles, by scheduling another guest application at lowest priority simultaneously with the main application on the node. Our experiments showed that by using this technique, the idle cpu cycles could be stolen by the guest application. However, it typically led to around 4-5% slowdown of the main application.

Date: Thursday, November 3^rd, 2005
Time: 3:00 PM
Place: 550-PGH



Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Jaspal Subhlok