GOVERNMENT

Microway, a leading provider of HPC solutions, announced the availability of the commercial version of MPI Link-Checker, A software diagnostic that finds underperforming nodes in MPI based clusters. MPI Link-Checker is the world’s first referential performance integrity tool that enhances cluster reliability and management, while lowering the TCO of scale-out computing. Initially available as a Beta release and free download on Microway’s website, the software has gone through extensive real-world testing on some of the world’s most demanding HPC installations. Based on customer feedback the new commercial version contains enhancements that make it possible to detect ‘hard to find’ problems in large clusters. The commercial version adds a powerful new off line data collection/analysis facility while retaining all the popular features from the successful Beta release, including the ability to check for bad cables, motherboards, NICs, switches, BIOS’s and OS’s, in real time. “MPI Link Checker,” commented Stephen Fried, President and CTO of Microway, “is the first product that makes it possible to diagnose intermittent cluster problems. HPC customers can now discover and highlight previously hidden faults like cables that are not properly seated or that have gone bad from physical abuse such as kinks due to re-routing or inadvertent damage. The product quickly pinpoints operational problems saving the cluster owner money and bringing the cluster back up to its full speed and operating potential.” Finding exactly what makes a node go ‘bad’ in a large cluster is a classic but non-trivial ‘needle-in-a-haystack’ problem. Using the off line data collection capabilities, MPI Link-Checker™ can probe the cluster and then analyze the collected data at a later date. The analysis is not trivial, as hundreds of megabytes of performance data can be collected on large clusters over a four day burn in period. “Sifting through this amount of “bad node data” quickly and intelligently requires the correct tool,” added Mr. Fried. “MPI Link-Checker™ now makes it possible to drill down into the analysis grids generated by large clusters. The new product can also dynamically view plots of transfer time and bandwidth vs. packet size for all the nodes in an analysis matrix, reduce analysis time by breaking large clusters into node groups and select the statistical method used to view the data. This last feature when combined with off line collection makes it possible to isolate intermittent problems that have heretofore been impossible to find! “ Ann Fried, CEO of Microway commented, “for the first time scale-out HPC cluster sites can drive overall performance in that we even enable them to spot problems within MPI itself. Often these problems are the result of inefficient device drivers or the wrong choice of parameters, such as the transition point between the Eager and Rendezvous protocols. Strengthened for commercial use and enhanced with more powerful features based on our successful beta program, we expect that the commercial version of MPI Link-Checker will become an essential tool for all MPI based Linux clusters.” A full description of the product is available at www.microway.com. Background: Incorporated in 1982, Microway is a major vendor in the High Performance Computing market, designing state of the art, high end Linux clusters, servers, and data storage solutions. Users worldwide pushing the limits of technology choose us for solutions. These include universities, life sciences, financial, military, Fortune 500s and research agencies. Microway partners with leading commercial software providers to include products such as SUSE and Red Hat Linux, Microsoft Windows Server 2003, Platform Computing LSF, Intel, PathScale and PGI Compilers, and MPI Software Technology MPI/Pro on its Opteron based clusters. Microway is an AMD Platinum Partner, Intel Premier Provider, Novell Gold Partner and Microsoft Direct OEM for Windows Server HPC License. Classified as a small business, woman owned and operated, Microway's GSA Contract Number is GS-35F-0431N. Trademarks include GigaCube, MCMS, MPI Link-Checker, Navion, NumberSmasher, NodeWatch, and Quadputer. For more information and a subscription to Microway's online technical newsletter, please visit www.microway.com