Mitrion Hybrid Computing Platform Delivers 20x Performance Increase and 90% Power Savings for Information Retrieval Application
Matrixware-Sponsored, Green Computing Project Performed by University of Glasgow Computing Science Department.
Mitrionics, Inc, the technology leader in FPGA-based hybrid computing, today announced that its Mitrion Accelerated Computing Platform achieved a 20x performance increase and ninety percent power savings over traditional processors when running standard application algorithms for Information Retrieval and document filtering. In a project sponsored by Matrixware Information Services and supported by Mitrionics, researchers from the University of Glasgow conducted comparative tests to measure the performance differences between running Information Retrieval algorithms on traditional processors versus Field Programmable Gate Arrays (FPGAs). To develop the tests, computational portions of the algorithm were rewritten using the Mitrion-C programming language, and then run on Mitrion Virtual Processors operating on Xilinx FPGAs contained within a supercomputing system designed by Silicon Graphics, Inc and provided by the Information Retrieval Facility. The complete study is available for download at: http://www.mitrionics.com/?page=ir
"Because of the time constraints on the project, we needed a rapid development environment. The Mitrionics SDK allowed us to develop a very complex FPGA application at a fraction of the time required for traditional HDL design, and in our case without any performance loss," said Wim Vanderbauwhede, PhD, EPSRC Advanced Research Fellow at the Department of Computing Science, University of Glasgow. "We were really pushing the boundaries of the SDK for this project, in particular with the high-speed direct I/O to the host processor, and the collaboration with the Mitrionics developers proved very productive.”
“FPGAs have huge potential to deliver tremendous speed-ups and power savings for many applications – so they are super green!” said Leif Azzopardi, PhD, RCUK Research Fellow at the Department of Computing Science, University of Glasgow. “And with Mitrion-C being such a high-level language it is a lot simpler and faster to implement sophisticated matching algorithms on the FPGAs than coding on bare metal.”
“Reconfigurable computing represents an important technology in the quest to develop high performance systems able to reduce the ever increasing energy drain by data centers,” stated Francisco E. Webber, CEO of Matrixware Information Services. ”The prototype IR filtering system developed by the University of Glasgow demonstrates a potential application to patent searchers for interactive online filtering and can be extended to other information retrieval fields. The potential benefits for the industry are huge.”
The full results, methodologies, and details of the study were presented by representatives from the University of Glasgow at the FPL09 Prague - 19th International Conference on Field Programmable Logic and Applications held in Prague, Czech Republic, August 31st through September 2nd, 2009.
Abstract: “FPGA-Accelerated Information Retrieval: High-Efficiency Document Filtering”, Authors: W. Vanderbauwhede, L. Azzopardi, and M. Moadeli from the Department of Computing Science, University of Glasgow.
Power consumption in data centers is a growing issue as the cost of the power for computation and cooling as become dominate. An emerging challenge is the development of “environmentally friendly” systems. In this paper we present a novel application of FPGAs for the acceleration of Information Retrieval algorithms, specifically, filtering streams/collections of documents against topic profiles. The results show that FPGA acceleration can result in speed-ups of up to a factor of 20 for large profiles. The complete study is available for download at: http://www.mitrionics.com/?page=ir
Mitrion SDK 2.0 - Accelerated Computing Platform
The Mitrion SDK is a complete development environment for accelerating applications while making it easy for programmers to learn how to write software using parallel programming techniques that take advantage of the parallel processing capabilities of the Mitrion Virtual Processor. The SDK includes a compiler and graphical code-simulator that simplifies finding programming errors, performance bottlenecks, and inefficient code. Mitrionics offers a free SDK Personal Edition that supports Linux/UNIX, Windows and Mac OS X which can be downloaded from: http://www.mitrionics.com/?page=Downloads
About Matrixware Information Services
Matrixware Information Services offers superior solutions and services for professional information retrieval to the global market. These solutions and services help organisations to face the information economy and, thereby, provide them with a distinct business advantage.
Matrixware builds strong, trusting relationships through cutting-edge, open science, open source and open business concepts. Matrixware’s capabilities are fuelled by the findings of leading global scientists. For more information, visit: http://www.matrixware.com/
Founded in 2001 and privately held, Mitrionics is the technology leader in FPGA-based application acceleration and hybrid computing. The Mitrion Platform, which includes the Mitrion-C Compiler and Mitrion Virtual Processor, utilizes parallel programming and parallel processing to enable greater processing performance and a greener computing alternative. The Mitrion Platform is unique because it eliminates the need for circuit design skills, thus making FPGA-acceleration accessible to scientists and developers all over the world. Mitrionics has key industry relationships with chip companies Intel, AMD, Xilinx and Altera, systems vendors HP, SGI, and Cray, and module suppliers Nallatech, XtremeData and DRC. For more information, visit the company Web site at http://www.mitrionics.com, email firstname.lastname@example.org or call 408-395-3247.