Mark Baker Tellez, director of business development for SRC, said the recently installed SRC-7 system is either the first or among the first large reconfigurable computing clusters to be deployed anywhere in the world.

The SRC-7 MAPstation cluster features a Stratix II EP2S180 FPGA from Altera Corp.

"We think this is one of the first large reconfigurable computing clusters to be deployed," Tellez said.

Jack Dongarra, a distinguished professor at the University of Tennessee's department of electrical engineering and computer science, said he couldn't verify that this installation was a first because he couldn't judge the size of the system from the information provided by SRC. Tellez said SRC was unable to provide detailed information about the system due to contractual obligations.

The SRC-7 system was installed in support of a joint research project between JSU and the U.S. Army Engineer Research and Development Center, according to SRC.

The system is described as the world's first Infiniband-based SRC-7 MAPstation cluster. Each of the system's nodes is said to have the performance of hundreds of microprocessors. The system consumes 90 percent less power than a microprocessor-based cluster and is capable of scaling to tens of thousands of nodes, according to SRC (Colorado Springs, Colo.)

How easy is it to program?
Sounds good, but how easy is it to program the system?

According to Tellez, the challenges in programming a reconfigurable system largely depend on the tools utilized and approach one takes.

"Our customers have equated the effort required to move their code to our reconfigurable system using the SRC's Carte Programming Environment as being on par with that required to convert a traditional uniprocessor application to an MPI version," he said. "Carte allows the developer to use true standard ANSI C or Fortran, not a 'C-Like or Fortran-Like language' to program the entire system. We find that most customers are pretty comfortable utilizing the system after attending a short introductory course (two days). If they know how to program in Fortran or C they really should not have any difficulty coming up speed very quickly."

According to Tellez, MAP-based systems provide more processing power per processor and consume far less power. A system with eight MAPs can have the performance of a microprocessor-only system with 800 processors, he claimed. Systems using MAP processors also consume 10 to 100 times less power per unit of computation than systems with microprocessors alone, he said.

In a statement issued last month about the deal, Khalid Abed, a faculty member in the department of computer engineering at JSU and the principal investigator on the research effort, said microprocessor-based high performance reconfigurable computing (HPRC) clusters are facing challenges in runtime performance, memory bottlenecks, floor space and power dissipation. The The SRC-7 system will allow Abed's team to investigate how to map complex parallel codes onto HPRC clusters, he said.

"Reconfigurable supercomputers allow researchers to reconfigure the hardware for different computing applications to achieve optimum performance for such applications instead of running different types of applications on a fixed supercomputer," Abed said.