Cover Story
ARL MSRC Increases its Sustained Performance to Over 100 HABUs
In early 2007, the Army Research Laboratory Major Shared Resource Center (ARL MSRC) will increase its computing capability from 27 to 108 HABU equivalents (one HABU equivalent represents the equivalent performance of the weighted DoD benchmark suite on a 1024 CPU IBM power 3 system) by adding four computing systems to its already robust computing spectrum The factor of four increase in capability will make the ARL MSRC one of the largest computing centers in DoD.
Four commodity-based supercomputers were procured for the ARL MSRC as part of the HPCMP's Technology Insertion 2006, an initiative to modernize DoD's high performance computing capabilities. The HPCMP provides the supercomputer services, high-speed network communications, and computational science expertise that enables the U.S. Defense laboratories, such as the ARL MSRC, to conduct a wide range of focused research, development, and test activities.
 |
"This increase in computing capability will give DoD scientists and engineers the ability to solve complex, three-dimensional, time-dependent, physics-based problems in a timeframe that can provide the data necessary to assist with weapon development and procurement decisions," said Charles J. Nietubicz, Director of the ARL MSRC and Chief of the High Performance Computing Division in the Computational and Information Sciences Directorate.
Leading Edge Clusters
Four Linux Networx Advanced Technology Clusters will be installed at the ARL MSRC this summer. The most powerful of the new systems is a 1024-compute node cluster with 4096 3.0 GHz Intel Woodcrest cores for computation. This system will increase the ARL MSRC's computational capability by more than 50 TFLOPs. The system will also have 112 3.0 GHz cores (28 nodes) for login, storage, and administration, and will have 8.5 TB of memory and 260 TB of disk. All nodes will communicate via a 4X DDR (20 Gbps) Infiniband network with 10 gigE uplink capability. This system will be ranked in the top 15 of the world's most powerful computer systems.
The second system, a 842-compute node Advanced Technology Cluster will comprise of 3368 mid voltage 3.2 GHz Intel Dempsey cores for computation. The system will increase the ARL MSRC's computational capability by more than 21.5 TFLOPS. The system will also have 96 3.73 GHz cores (24 nodes) for login, storage, and administration and will have 7.1 TB of memory and 260 TB of disk. All nodes will communicate via a 4X DDR Infiniband network with 10 gigE uplink capability. This system will also be ranked in the top 15 of world's most powerful computer systems.
Tom Kendall, Chief Engineer of the ARL MSRC said "The combination of dual core processors with dual memory busses, DDR 4X Infiniband and fully buffered memory into a power optimized system will provide DOD researchers with an unprecedented level of high performance computing capability and capacity".
The third system is a 64 core test and development system comprised of 8 compute nodes with Dempsey cores, 8 nodes with Woodcrest cores and 7 TB of disks to be delivered in advance of the other two systems. The Linux Networx and ARL MSRC teams will work closely utilizing this machine for DoD kernel and applications testing. "Interconnects and file systems are critical to what we do, ironingout these details on a smaller machine is much easier. We want our users to have a positive first experience when they get onto the larger production machines" adds Tom Kendall.
Finally, the ARL MSRC announces the first 64-node scientific visualization supercomputing cluster in DoD with dual nvidia Quadro FX 4500 cards per node and an Infiniband network. This supercomputing cluster is also from Linux Networx.
Visualization Cluster
The visualization supercomputing cluster will allow for interactive visualization of enormous datasets. By performing both the visualization algorithms and rendering in parallel a scalable solution for analyzing calculations consisting of billions of computational cells will be achieved.
"The complexity and fidelity of current production DoD calculations consume terabytes of memory and tens of CPU years for a single simulation" said Jerry Clarke, Scientific Visualization Team Lead. "When you're dealing with that amount of data, every aspect of visualization must be parallel. This system will allow for the interactive analysis of important calculations currently impossible on available resources."
 |
| Partners for TI-06 Success: Initiative to Build the Largest DoD Clusters in 2006 |
What is a HABU?
According to www.habu.org, a habu (pronounced "hah-BOO") is a poisonous snake found in southeast Asia (Japan, Phillipines, Taiwan, southeast China). Habus are pit vipers, more closely related to the adder than to any species of North American snake.
Habu is also the nickname of the SR-71 Blackbird aircraft. The name was given to the plane by the people of Okinawa because the strange shape reminded them of the snake.
In computing terms, one HABU equivalent represents the equivalent performance of the weighted DoD benchmark suite on a 1024 CPU IBM POWER3 system. Brainerd, an IBM POWER3 came into production at the ARL MSRC in 2001 is equivalent to 1 HABU. This Fall the ARL MSRC is replacing it with a commodity cluster equivalent to 52.2 HABUs which will occupy the same footprint providing a factor of 50 improvement in 5 years.