Welcome to the Research Computing Subcommittee site!

The Advisory Committee on Research Computing (ACRC) also serves as the Research Computing Subcommittee of ITSPAC. The ACRC is chaired by Iwan Alexander, who also serves as the lead on the Research Computing Subcommittee. The ITS staff lead of the subcommittee is Roger Bielefeld, who also serves as a member of ACRC.

The subcommittee was charged with identifying its three highest priority projects for FY 2006-2007 for inclusion in the ITS strategic plan. The subcommittee was to produce a proposal for each project using the standard Budget Request Proposal (BRP) form no later than 15 March 2006.

The subcommittee met on 13 February 2006 and discussed three projects, which were then described at the 14 February 2006 ITSPAC meeting. Subsequent discussion occurred and the project list was refined and approved. The subcommittee finalized the necessary BRP documents and submitted its list of highest priority projects on 15 March 2006.

The subcommittee identified its three highest priority projects for FY 2006-2007 as outlined below.

[edit] Expanded Pilot High Performance Computing Facility

The ITS pilot HPC cluster was ordered in October/November 2004, became operational in March/April 2005, and was made available to users in May 2005. The pilot cluster consists of 18 compute nodes (36 processors), 1 master node, and 1 management node. Each compute node consists of two processors, 4G of memory, and a hard drive. Applications software can run in a single processor, in a single compute node utilizing both processors, or in several compute nodes simultaneously in a distributed fashion.

Since May 2005 usage has grown to 31 users from 19 research groups based in 11 different departments (Astronomy, Biology, Chemistry, Electrical Engineering and Computer Science, Geological Sciences, Mathematics, Mechanical and Aerospace Engineering, Pediatrics, Physics, and Statistics). During most of this time (until mid-February), the software offered on the cluster was limited to C, C++, and FORTRAN compilers, and libraries for numerical computation.

Funding was recently obtained earlier this year for licensing additional software on the cluster, and there is now an effort underway to install both commercial and open source software to support general computational and visualization work (Mathematica and MATLAB), computational fluid dynamics (FLUENT and FIDAP), computational chemistry (GAMESS, Gaussian, and Amber), statistical modeling and visualization (R), database research (MySQL), and other disciplines. The addition of this software will attract usage by additional faculty in these and other departments. In particular, significantly increased usage is expected from Biomedical Engineering, Chemical Engineering, Chemistry, Electrical Engineering and Computer Science, Materials Science, Mechanical and Aerospace Engineering, Physiology and Biophysics, Radiology, and Statistics.

In order to provide for this expected increasing demand on HPC resources, the subcommittee recommends that the pilot cluster be enhanced in three ways: (1) increasing the number of compute nodes (and processors) in the cluster, (2) adding at least one compute node having at least eight processors and at least 16G of memory, and (3) add at least 2T of disk storage to serve as scratch storage for the cluster.

Increasing the number of processors in the cluster permits a higher degree of parallelism to be supported in parallel computational algorithms. It also permits a larger number of different computational tasks to run simultaneously.

Adding a compute node having a large number of processors and a large amount of memory shared among those processors permits parallel algorithms in which processors use memory to frequently share information. Such algorithms are not well-supported in a cluster having its memory distributed among the individual processors, with inter-processor communication supported by a communications network such as Gigabit Ethernet or InfiniBand. In order to balance the pilot HPC resources that are available at Case, the subcommittee proposes installing a modest-size shared memory (SMP) computer as a special node in the cluster for use by applications requiring a shared memory architecture for optimal performance.

Adding at least 2T of disk storage as scratch storage for the cluster will serve the needs of some applications that generate and utilize large amounts of data. This storage implementation must be designed to permit easy and inexpensive expansion to support researchers who require large amounts of permanent online storage.

[edit] Add staff member to support the high performance computing initiative

Since its inception, the administration of the ITS pilot cluster has been primarily outsourced to OSC. While expertise in HPC certainly exists at OSC, from the Case point of view there are some aspects of current support that are lacking. First, OSC support requires adherence to the OSC standard operating procedures for HPC, which may not be suitable for the needs of Case researchers. Second, OSC may not be able to support the installation and operation of software applications that are not used at OSC. Third, OSC support is provided against the background of OSC activities that may interfere with service requests from Case. Fourth, support is limited to "behind the scenes" support of the cluster with direct contact with Case limited to a handful of ITS staff.

The subcommittee recommends the addition of a new staff position to support HPC at Case. Such a position would permit increased flexibility in the operations of the Case HPC facility by allowing divergence of operations from OSC standards. It would permit the acquisition of more open source software that the new position would be able to install and maintain. It would increase the responsiveness to faculty needs related to cluster configuration and to operational needs related to software installation and update. It would permit alignment of priorities more closely with Case needs rather than OSC needs. Finally, it would increase direct support to faculty researchers as this position could have limited contact with faculty when appropriate.

[edit] Continue the matching funds program offered by the Provost and the VPITS/CIO

In the fall of 2004 the Provost and VPITS/CIO announced a program of matching funds that permitted researchers with external grant funding to multiply their investment in equipment and services related to the central research computing facility. The Provost and VPITS/CIO extended this program into the 2005-2006 fiscal year. The subcommittee believes that interest in this program exists among the faculty, but many faculty have not participated in the program for a variety of reasons:

Along with the continued availability of these funds, two additional actions by ITS are required to improve the chances that faculty would consider applying for them. First, a well-defined description of the matching funds program needs to be distributed. Second, very strong support to faculty in creating budgets that incorporate the program must be made available so that they can more effectively compete for funding opportunities such as the NSF CRI program (http://www.nsf.gov/pubs/2004/nsf04588/nsf04588.htm) and the NSF MRI program (http://www.nsf.gov/pubs/2005/nsf05515/nsf05515.htm).