MRI: Acquisition and Analysis of a Multi-Disciplinary Beowulf Cluster

点击翻译按钮获取中文摘要

OCI 0722819Calvin CollegeJoel C. AdamsMRI AcquisitionAt tier-1 research universities, the trend is to build massively parallel clusters of hundreds or thousands of nodes, connected using a low-latency interconnect like Myrinet or Infiniband. These clusters provide excellent performance, but their expense puts them out of reach of the hundreds of small universities, colleges, and community colleges across the U.S. (and world).By contrast, the Beowulf cluster [1] is a high performance computing platform consisting of a modest number of PCs (usually 48 or less), an inexpensive network (usually TCP/IP), an open source operating system (usually Linux), and parallel programming software (usually MPI). By trading off performance against price, the Beowulf cluster's low price makes it affordable at small universities, colleges and community colleges that cannot afford a large cluster.This proposal is to build a 32-node Beowulf cluster for multi-disciplinary research and teaching at Calvin College, a non-PhD granting comprehensive college of about 4200 students and 310 faculty. According to NSF data, Calvin is first in the U.S. among baccalaureate and masters granting institutions for producing undergraduates who complete their PhD in computer science.According to NRC data, Calvin is fourth in the U.S. among such institutions for producing undergraduates who complete science PhDs. More than 25% of their students major in the sciences. The Department of Computer Science and the Department of Engineering offer ABET-accredited programs; the Department of Chemistry and Biochemistry is ACS-accredited.