Research Computing History - Summary

    Research computing at SDSU traces its history back to the 1960's.  At that time, the  Agricultural Experiment Station was instrumental in helping lease  an early IBM Computing System for academic and research use.   This began a progression of IBM mainframes that were ultimately used by an increasing number of research groups on campus.  During this period  research computing was managed by a  Director of Research and Data Processing.  As the College of Engineering and the Remote Sensing Institute began to have unique requirements,  the data center (as well as the  units themselves), installed dedicated systems for their needs.  These progressed thru a series of Prime Computer Systems and IBM RS/6000's.  

    In 1991, SDSU received funding assistance from the Governors Office of Economic Development to help purchase an IBM ES/9000 Vector Processor.  Along with the purchase,  IBM provided vector processing training for a mechanical engineering graduate student and  two Computing Services staff members.  This was SDSU's first foray into HPC parallel processing.  

    In the late 90's, SDSU consolidated numerous IT groups on campus under the auspices of a CIO.  Since that time, the research computing function has   been paired with several IT groups until its current position within the University Networking group.  

    In 2000, the university did a large scale upgrade of the campus network.  In 2002 there was a mass upgrade  of  personal computers on the campus.  During this period, Computer Services staff were beginning to test various Linux distros.  The choice of SUSE for our systems was largely a result of our early Beta testing of SUSE Linux on our mainframe systems.  With increasing experience with Linux and the availabilty of older switches and personal computers, Computing Services built it's first Beowulf Cluster out of surplus Gateway desktops.  In 2002 we deployed a 17 node cluster for research use.  

    In late 2001 there was a need for increased central storage by several IT groups, including research computing.  A report was written for the CIO and a plan for SDSU's first SAN was put together.  Two systems were chosen.  A Compaq Storageworks system and an IBM Shark were   installed.  Centralized research data was located on the Shark system.  These systems were maintained until storage requirements outgrew them.   Since that time storage has been consolidated onto  a Hitachi AMS platform.  

    In late 2003 SDSU received an EPSCOR grant to purchase an IBM 1350 cluster.  The initial cluster was installed and brought online in early 2004.  Initially the cluster contained a management node plus 19 dual processor AMD compute nodes.      The interconnect was 100 Mb ethernet and node memories were 2 GB.   In following years this cluster was upgraded in several iterations using a combination of SDSU funds and IBM Matching Grants.   The final configuration had 26 compute nodes.  The system was eventually transferred to the SDSU Computer Science Department for academic use.  

    In 2003 SDSU also received EPSCOR money to build a visualization environment using Geowall technology.  Five systems were built and deployed using IBM Intellistation workstations.  

    In the summer of 2007 SDSU purchased our second cluster.  This one was built with an IBM Bladecenter and dual core AMD processors.  The initial configuration had 10 compute nodes (40 cores) plus  a management node.  The interconnect was via Gig ethernet and memory was 4GB.    This cluster has also been  upgraded several times with combinations of SDSU funds and IBM Matching Grants.  The final configuration has 42 compute nodes with 316 cores and 8 to 32 GB memory.  Cluster2 used Torque/Moab for resource management and scheduling, SLES for operating system and CSM for cluster management.

    In 2008  Research Computing began a move toward providing more domain specific support.   The initial expertise and focus was within Chemistry.  The addition of this support was instrumental in increased campus support for the  HPC environment.   Currently there is added emphasis on  providing more domain support within the life sciences.  

    In 2010-2011 SDSU applied for and received an NSF Research Infrastrucure grant that, among other things, allowed us to add a 10 Gb ethernet core within the the data center.  This gave our research servers  10 Gb ethernet interconnections.  

    In 2011 SDSU purchased its current IBM iDataPlex cluster.  This was purchased in two phases and contains a managment node, 71 compute nodes (852 cores) and 18 GPU's.  The interconnect is a combination of Gig ethernet and  QDR infiniband.  Memory is 48 to 96 GB per node.   Bigjack uses Torque/Moab for resource management and scheduling functions.  With Bigjack we have transitioned from CSM to XCAT for cluster management and provisioning.

    Various options are currently being investigated to improve the environment for research computing on campus. Areas of interest include more domain level expertise, large scale memory aggregation, parallel file systems, higher speed data transfers, heterogeneus processor environments, long term archiving, visualization, viable uses of virtualization,  etc.

    Tag page (Edit tags)
    • No tags
    You must login to post a comment.
    Powered by MindTouch Core