Parallel computer clusters now represent a mainstream resource of commodity compute cycles in many academic departments. We review our experiences over the past 7 years of building and experimenting with compute clusters and speculate about future trends. We focus on scheduling and utilisation and discuss data collected from a complete year of use of one system. We discuss scheduling inefficiencies and the lost or missing cycles that are inevitable when running a multi user resource.
Keywords: scheduling and resource management; recent historical trends in cluster computing.
Full Document Text: PDF version.
Citation Information: in Proc. Parallel and Distributed Processing Techniques and Applications (PDPTA'05) June 2005, Las Vegas.
BiBTeX reference:
@inproceedings{CSTN-018,
Address = {Las Vegas, USA.},
Author = {K. A. Hawick and H. A. James},
Booktitle = {Proc. Int. Conf on Parallel and Distributed Processing Techniques and
Applications (PDPTA'05)},
Month = {June},
Title = {Trends in Cluster Computing Scheduling and the Missing Cycles},
Year = {2005},
series = "CSTN-018"
}
\bibitem{CSTN-018}
H. A. James and K. A. Hawick,
Trends in Cluster Computing Scheduling and the Missing Cycles,
Technical Note CSTN-018, and in:
Proc. Int. Conf on Parallel and Distributed Processing techniques and
Applications (PDPTA'05), June 2005, Las Vegas, USA.