October 25, 2011
Last month Rensselaer Polytechnic Institute (RPI) announced it had been awarded a $2.65 million grant to acquire a 100 teraflop Blue Gene/Q supercomputer for its Computational Center for Nanotechnology Innovations (CCNI). The new system will also include a multi-terabyte RAM-based storage accelerator, petascale disk storage, and rendering cluster plus remote display wall system for visualization.
Even though the yet unnamed Q machine is just a microcosm of a true petascale supercomputer, it is designed to be used for exascale research: scaling codes, exploring alternative approaches to checkpointing, and dealing with I/O bottlenecks. The supercomputer will also provide a home for a variety of research applications at Rensselaer.
According to the press release these projects include: "developing new methods for the diagnosis of breast cancer using data from non-invasive techniques; modeling plasmas to aid the design and safety of future fusion reactors; modeling wind turbine design to increase efficiencies and reduce maintenance; application of new knowledge discovery algorithms to very large semantic graphs for climate change and biomedical research, modeling heat flow in the world's oceans, integrating data and computations across scales to gain a better understanding of biological systems and improve health care; and many others."
This is the first machine CCNI will deploy with NSF funding behind it and the first new supercomputer at the center since it launched five years ago. CCNI was kicked off in 2006 with a $100 million investment from New York State, RPI, and IBM, using the initial cash to build out the center, hire staff, and acquire HPC resources. Its stated mission: to advance the science of semiconductor manufacturing and related nanotechnology applications for academia and industry.
The NSF money to buy the Blue Gene/Q system came out the agency's Major Research Instrumentation Program, which, as the name implies, funds instruments for scientific and engineering research. These include devices such as mass spectrometers, X-rays, laser systems, microscopes, as well as a variety of computational resources. Because of NSF's involvement, time on the system will be available to researchers nationally. Rensselaer scientists and engineers, as well as those at other New York state universities will also be able to bid for cycles on the system.
The first Rensselaer supercomputer was a Blue Gene/L system, along with a Power-based Linux system and some smaller AMD Opteron clusters. The Blue Gene/L system, which is still operational, delivers 90 teraflops and represents most of the computation capacity at CCNI. When installed in 2007 it was the seventh most powerful system in the world. Despite CCNI's rather modest computational capacity by 2011 standards, more than 700 researchers spread out across 50 universities, government labs, and commercial organizations have used the center's HPC resources to run their science and engineering workloads.
Although the upcoming Blue Gene/Q is relatively small as supercomputers go -- a mere 100 teraflops -- it will provide as much computational horsepower as the older L system plus all the remaining clusters at the center, According to CCNI, the upcoming system will fit into just half a rack -- about 1/30 the space as center's original Blue Gene machine.
And, because it's a Blue Gene Q, it should provide some of the best performance per watt on the planet. A similar 100 teraflop Blue Gene/Q prototype system, which is housed at IBM's T. J. Watson Research Center, delivered 2097 megaflops/watt (the number one system on the latest Green500 list), and consumed just 41 KW. To put that in perspective, the 2005-era ASC Purple supercomputer also delivered 100 peak teraflops, but consumed a whopping 7,500 KW.
According to CCNI Director James Myers, for the time being will keep their other HPC systems, including the Blue Gene/L, operational. But he admits that it will probably make sense at some point to decommission the older machines, considering how little performance per watt they are delivering. In general, the operational costs of maintaining five-year-old HPC machines these days is often better spent on adding newer, more energy-efficient capacity. "We are certainly paying attention to those lifecycle costs," says Myers.
The new Blue Gene/Q system is scheduled to be installed in 2012, in the same general timeframe that Argonne and Lawrence Livermore National Labs are expected to deploy their much larger Q machines: the 10 petaflop "Mira" system and the 20 petaflop "Sequoia," respectively.
It is also designed to be a platform for data-intensive applications. The RAM-based storage accelerator that is to be integrated into the system will be a critical component for data-intensive research. Essentially the accelerator is a 2-4 terabyte RAM disk that will be used to greatly speed up I/O for disk-bound applications. It will also be used to support interactive visualization by streaming data from the RAM disk to the visualization cluster without going through the bottleneck of disk storage. According the Myers, the RAM disk is to be based on commodity components, although its exact makeup is still to be worked out.
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?