DOE flexes computing muscle
Connecting state and local government leaders
Energy's Pacific Northwest National Lab has deployed a Cray XMT supercomputer to accelerate data analysis and predictive analytics in energy, national security and fundamental science.
The Energy Department's Pacific Northwest National Laboratory (PNNL) has deployed a Cray XMT supercomputer to accelerate data analysis and predictive analytics in the areas of energy, national security and fundamental science.
An early release system has been installed in PNNL's Applied Process Engineering Laboratory in Richland, Wash. The supercomputer will be upgraded to a full production system in 2008, lab officials said.
The system also will be available to researchers at Washington State University, increasing opportunities for collaborative science and discovery.
'I expect the new system will provide researchers with levels of performance for data analysis that they have never achieved before,' said Moe Khaleel, director of computational sciences and mathematics at PNNL.
PNNL researchers and colleagues at Cray have obtained promising results in testing the system's massively multithreaded architecture on several PNNL-developed applications, including one that constantly monitors the state of the power grid, PNNL officials said.
Cray XMT's architecture and large global memory are configured for applications ' such as data discovery, business intelligence, bioinformatics and power grid analysis ' that require access to terabytes of data arranged in an unpredictable manner.
Conventional computer systems are stymied by the random and unpredictable data access required by these applications. Performance in traditional systems is determined mainly by the speed at which the memory can deliver information to the microprocessor. However, in many large-scale applications, the processor is idle 80 to 90 percent of the time while waiting for data from memory.
The Cray XMT system is designed to overcome memory bottlenecks because each processor can handle 128 independent threads. The system can scale to hundreds of thousands of threads in a shared memory. Threads are portions of a program that can run independently of and concurrently with other portions of the program, company officials said.
NEXT STORY: From Earth to the moon