On September 15, 2008, UIT went live with its new high-performance computing research cluster (http://go.tufts.edu/cluster). With the arrival of the new cluster, it was time to retire the old one. But instead of throwing it away, UIT decided to give it away to a Tufts research group that was capable of supporting and maintaining it.
In January 2009, UIT put out a call for proposals to the Tufts research community on new uses for the old cluster. Interested parties were asked to submit a one page proposal that described:
- where and how the old cluster would be housed.
- who would maintain and support the old cluster in the new environment.
- the research project that would use the old cluster.
|Douglas Hwang E09 and Jason Prapas E11 tour the Tufts data center in preparation for a project they’re working on. By Joannie Tobin/Tufts Photo.|
Four competitive proposals were submitted to a five-member review committee composed of UIT staff and non-UIT faculty. Chaired by Bruce Boghosian of the Mathematics department, the committee also included professors Soha Hassoun and Norman Ramsey of Computer Science and UIT Academic Technology associate directors Gina Siesing and Lionel Zupan.
After reviewing and ranking the proposals based on specific evaluation criteria, the committee selected a proposal submitted by the department of mechanical engineering entitled, “Validation of Power Consumption Models for Air and Liquid Cooling on the Dell 2600 Cluster.” The authors were Marc Hodes, Associate Professor of Mechanical Engineering; Vincent Manno, Professor of Mechanical Engineering, Rich Wlezien, Professor of Mechanical Engineering; and Robert Hanneman, Director of Tufts Gordon Institute and Professor of Practice in Mechanical Engineering.
This group of researchers proposed using the old cluster not for computation per se but to address the problem of high energy consumption by data centers in the United States. Data centers now account for 2-3% of electricity consumption in the U.S. Approximately half of the power supplied to a data center drives the computational hardware, while the remainder of it (i.e., about 50%) is consumed by the cooling equipment necessary to dissipate the heat produced by it.
Since energy level models do not exist that adequately address this problem, the group is developing system-level tools to enable the analysis of data centers as macroscopic thermodynamic systems. This will allow the assessment of design options that connect the energy supply and consumption, and the building of environmental systems.
Thanks to all who participated and congratulations again to Marc Hodes and his team for their ingenious proposal!
Rebecca Sholes, Senior Faculty Development Consultant, UIT Academic Technology