If you had the choice of High-Performance Computing (HPC) or low-performance computing, which one would you pick? Nobody asks if you want low-performance computing, because who would want that? Maybe if you didn’t care when you got the results, or weren’t trying to be competitive, or the first to discover or invent something, then “regular” computing would do. We hear that high-performance gasoline might be over killing for most car engines, but we’ve been brainwashed to think our car engines need the higher octane (performance) gasoline.
So what’s up with HPC? Is it worth the premium price? What makes it so special? Is performance truly better?
“Recognizing the strategic importance of growing our research capabilities has become a major goal of the University”
It’s interesting to be involved with HPC and to watch how it is transforming so many industries and disciplines. What previously was used strictly for research is now becoming table stakes for many corporations to stay competitive. Employers want graduates with strong computational and data analytical skills. And who isn’t tired of hearing about big data? Is anyone working on little data?
As the Chief Information Officer (CIO) of Southern Methodist University (SMU), I have had the opportunity to watch our HPC interest, investment, involvement and results continue to develop, mature and pay dividends.
A nationally ranked private university, SMU is a distinguished center for teaching and research located near the heart of Dallas. SMU’s 11,000 students benefit from small classes, research opportunities, leadership development, community service, international study and innovative programs. The University offers a strong foundation in the humanities and sciences and undergraduate, graduate and professional degree programs through seven schools.
HPC at SMU has significantly evolved over the past eight years. Recognizing the strategic importance of growing our research capabilities has become a major goal of the University. With limited wet lab expansion options, we concluded computational research was the best path forward for research at SMU. An initial grant proposal award from the National Science Foundation (NSF) brought funding for our very first HPC cluster. We also used some good old American ingenuity by leveraging some end-of-life PCs. What started with a request for a couple of old computers, turned into over 100 PCs linked together as a single cluster with enormous use for the biology professor that built it. Probably not the most advanced solution, but a very handy one to get started on a shoestring budget. Then we received an older Department of Defense (DoD) cluster with 8,800 cores–if we could get it shipped from Maui to Dallas! Did I mention that the 1,104 servers and 1.4 petabytes of storage arrived without hard drives because the DoD required the drives be destroyed? Hard lesson learned with obtaining a re-cycled cluster, but still it was worth the investment. Originally it was a $7 Mn cluster, Maui asked to buy it back from us. SMU students participated in the name-the-cluster contest. Combining our Mustang and pony heritage with our mainframe ancestry, the winning name ManeFrame got one student a new laptop! Millions of experiments representing many disciplines have run on this third-generation machine, but it has started to show its age with failing hardware and unavailable replacement parts. It was time to plan for a replacement cluster.
For the next generation cluster configuration, the director of our Center for Scientific Computing gathered representatives from each discipline to determine the proper number of worker nodes, high-memory nodes, graphics processing unit nodes and storage and networking requirements. We wanted something comparable to the 10,000-core cluster we would retire. And then there was the challenge of funding this replacement. A common challenge for universities is sustaining the equipment lifecycle replacement costs of these complex and expensive clusters. Having support from top administration was essential, and fortunately for us, our president, provost, dean of research, Chief Financial Officer, and the board of trustees all saw the value of HPC, and supported a strategic and significant investment in our next-generation cluster.
With the backing of the University’s leadership, and new configuration in hand, we were ready to select a vendor, purchase and install our new cluster. This sounds simple, but many variables need to be considered. Having the appropriate facility to house the cluster is a key starting point, and SMU is fortunate to have a brand new state-of-the-art data center. Telecommunications infrastructure to connect the campus and connectivity to the world also is essential. With a new fiber optics infrastructure connecting all 100 buildings to our new data center, and large redundant “pipes” to the Internet from our Internet Service Provider (Lonestar Education and Research Network), our plumbing was in excellent condition.
As with any large purchase, we partnered with our purchasing department to help with the Request for Proposal (RFP) process, developing a detailed timeline with critical milestones. The RFP was released, and soon vendors responded. Given the complexities of the request, we knew we’d have to be prepared to assess the various solutions and costs provided in the responses.
Our timeline from RFP process to full roll-out of the new cluster was eight months. It seems like a long time, but much needed to get accomplished from design to execution.
I can’t finish the story yet because we’re still in the process of procuring the cluster. But I can tell you that the faculty, researchers and students are extremely excited about SMU’s commitment to research through HPC. We truly believe our latest investment in HPC will continue to help achieve our goal of “World Changers Shaped Here”!