Contributing to the UMD High Performance Computing Environment

Benefits of Contributing to the UMD HPC Environment

The Deepthought cluster has grown using a "co-op" model design. Researchers who want to purchase HPC hardware can contribute to this existing cluster instead of setting up and managing a new HPC cluster. Contributing researchers receive high priority time allocations on the cluster proportional to the amount of CPU time their contribution adds to the cluster. While not all projects are well suited to this model, it has several advantages where the fit is good:

Professional Management: The purchase of equipment is often the easiest part of having a working HPC cluster. HPC clusters are complex beasts, and we have seen many departments and/or research groups purchase equipment only to have it languish unused as they did not adequately anticipate the issues in setting it up. The Division of IT staff will support your hardware contribution in our existing Deepthought cluster, and deal with system administration, performance tuning, and maintenance.
Increased scalability: If you buy a 10 node standalone cluster, you can never run your code on more than 10 nodes. By participating in the Deepthought2 cluster, even though you only contributed 10 nodes, you can leverage the larger size of the cluster to run much larger jobs. This will consume your time allocation faster, so you can only run half a month at double the size of your contribution. But this gives you more flexibility in your jobs, and could be useful in determining whether your jobs would benefit from greater parallelization.
Increased time flexibility: Similar to the previous point, if you bought a 10 node standalone cluster, you cannot run more jobs at one time than fit on those 10 nodes. If you have a major conference in the middle of the month, you might like the ability of the Deepthought cluster to allow you to use most of your time allocation in the first half of the month, even if that restricts your allocation for the remainder of the month.

In addition to the hardware contributed by researchers, the Division of IT has contributed significant compute power as well as storage and other support hardware over the years. This collaboration has made the cluster a valuable campus resource.

The cluster is managed by Division of IT staff, under the high level direction and guidance of the Allocations and Advisory Committee (AAC), whose membership is open to contributors to the cluster. In addition to making policy decisions regarding the cluster, the AAC is responsible for allocating time from the Division of IT's contributions to the cluster. Researchers on campus, whether part of contributing units or not, may submit applications requesting time on the cluster for specific projects.

The AAC grants allow researchers to explore using parallel computational methods in their research without the expense of purchasing a cluster. Sometimes, the project would benefit from a modest amount of HPC use, but not enough to warrant purchasing a cluster. Sometimes, you are unsure of whether the computational problems can benefit at all from HPC techniques, and you wish to investigate before committing funds. The AAC will grant time to projects it feels can benefit from and will make good use of this campus resource. There are no charges associated with this, and all we ask is that the cluster be acknowledged in any paper resulting from its use.

How to contribute to the clusters

The Allocations and Advisory Committee (AAC) and the Division of Information Technology can help. Send email to the Division of Information Technology to let us know of your interest . We will discuss your research requirements with you and work with AAC members to determine whether the cluster or high-performance computing in general is appropriate for your type of research. Test allocations are also available to help determine whether and how much your application would benefit from running in a high-performance computing environment.

Once everyone is in agreement that that an investment in the cluster makes sense, we will initiate discussions with the AAC and cluster administrators to iron out the specifications and associated costs of the hardware contribution.