Michigan Academic Computing Center

On 20 January 2006, the Board of Regents approved construction of the Michigan Academic Computing Center (MACC), a new 10,000 square foot data center in the Michigan Information Technology Center building. This facility, a partnership among the University of Michigan, Internet2, MAV Development, Merit Network, and the Michigan Information Technology Center Foundation, will provide the University of Michigan with 8,500 square feet for hosting a mix of computing platforms. The remaining 1,500 square feet of space will be used by the other partners. The Provost’s Office has responsibility for the facility, which will be managed by Rene Gobeyn, the Data Center Coordinator. It is tentatively scheduled for occupancy in March 2007.

Philosophy

The facility will host a mix of systems across its two sides, with the west side designed for high density systems and the east side for less demanding systems. It’s expected the west side will be the location for high performance computing clusters and other systems with significant power and cooling needs. File servers, web servers, etc. may be located on the east side. This drawing, courtesy of Luckenbach/Ziegelman Architects, shows the layout of the floor, with “1” indicating the east side and “2” the west. See also a current view of the center from the south.

The center is designed to be a cost-effective place to house high density platforms in particular, which often stress the power and cooling capacity of campus buildings. Therefore, the Provost’s Office will cover UM’s share of the construction costs and provide an ongoing operational subsidy. This investment is intended as an incentive to work cooperatively as a campus on populating the center with existing platforms and to perhaps co-invest in shared data and compute capacity in the future.

Definition of High Performance Computing

High Performance Computing, or HPC, refers to computation- and data-intensive processing that is typically done in support of research. There is no strict benchmark for HPC platforms, but these systems have power and cooling requirements that exceed those of traditional web servers, email servers, and local area network-oriented file servers. HPC clusters, for example, can consist of hundreds of server platforms placed in dense rack configurations and organized to work together on significant computational challenges. HPC storage systems are designed to scale to tens of terabytes or even petabytes of data.

Services

The center will be run primarily in a “condo model.” That is, end users will provide their own racks, servers, and distribution network equipment (consistent with policies and standards). The facility will provide UM tenants floor space for approximately 254 racks (not counting network distribution racks), generator-backed power, cooling (handling an average load of 200 Watts/square foot), basic network services, and the management services of the Data Center Coordinator.

Costs

As noted above, the Provost’s Office will cover UM’s share of the construction costs. An ongoing subsidy from the Provost’s Office will go toward rent and the Data Center Coordinator position. Power (for operating computers and the cooling equipment) will be metered and the costs passed through to data center occupants. Charges for attachment to the campus network, which will be extended to the data center, will be built into the base rack rental rates. Beyond providing the capital equipment of racks, servers, and network distribution hardware, end user tenants will be responsible for their own systems administration staffing. Details of the overall budget, use charges, and the subsidy level are still under development.

Information Sharing

Updates about the center will be sent to the joinable list <MITC.Data.Center.Info@umich.edu>. You can add yourself to the list by going to the UM Directory, searching for MITC.Data.Center.Info, binding (authenticating) yourself with your uniqname and Kerberos password, and then clicking on the “Join” button.

Operating Agreement

Since February 2006, working groups have been addressing the construction process and the operating policies and procedures for the MACC. The operating agreement documents the partnership between UM and MITC and is available for review. The document, available below, provides a framework for how the facility will be managed and how duties will be divided, how decisions will be made about which clusters and servers are appropriate for the facility, how access and security will be provided and monitored, and how operating costs will be measured and allocated.

As UM has never run such a shared facility, we are developing entirely new policies and procedures. Please direct all comments and questions to <UM.HPCC.Operating.Agreement@umich.edu>. You may not get a personal response to each comment, but all comments will be posted, unattributed, on this web site.

Future Opportunity

Beyond the strict condo model of space use, this facility provides the campus an opportunity to co-invest in shared compute and data capacity (e.g., a large cluster or a shared memory machine) and services (e.g., storage and backup). Such an initiative would provide the potential benefits of further cost savings, inter-disciplinary collaborations, and greatly enrich the research capabilities of the University. It can come to fruition through multilateral conversations among the colleges. Stay tuned for further campus communications on this topic if it’s of interest to you.

 Materials Available:

22 December 2006 – Draft UM/MITC MACC Operating Agreement (22kb PDF)

Michigan HPC Facility (Powerpoint presentation made at ACSF on 12 September 2006.)

Data Center Cost Categories (PDF of handout to ACSF members on 12 September 2006.)

Feedback on the Operating Agreement received by 17 August 2006. Please note that section numbers refer to an earlier draft of the Operating Agreement and will differ in the current draft.