Multithreaded supercomputer seeks software for data-intensive computing

The newest breed of supercomputers have hardware set up not just for speed, but also to better tackle large networks of seemingly random data. And now, a multi-institutional group of researchers has been awarded $4.0 million to develop software for these supercomputers. Applications include anywhere complex webs of information can be found: from internet security and power grid stability to complex biological networks.

The difference between the new breed and traditional supercomputers is how they access data, a difference that significantly increases computing power. But old software won't run on the new hardware any more than a PC program will run on a Mac. So, the Department of Defense provided the funding this month to seed the Center for Adaptive Supercomputing Software, a joint project between the Department of Energy's Pacific Northwest National Laboratory and Cray, Inc, in Seattle.

"The system will allow much faster analysis of complex problems, like understanding and predicting how the power grid behaves -- one of the most complex engineering systems ever built," said Moe Khaleel, director of Computational Sciences and Mathematics at PNNL, which is leading the project.

Other researchers in the software collaboration hail from Sandia National Laboratories, Georgia Institute of Technology, Washington State University and the University of Delaware.

These new machines are built with so-called "multithreaded processors" that enable multiple, simultaneous processing compared with the linear and slower approach of conventional systems. The Center will focus on applications for the multithreaded Cray XMT, one of which Cray delivered to PNNL in September 2007.

"Traditional supercomputers are not well suited for certain kinds of data analysis, so we want to explore this advanced architecture," said PNNL computational scientist Daniel Chavarría.

In previously published work, PNNL computational scientist Jarek Nieplocha used a predecessor of the Cray XMT to run typical software programs that help operators keep the power grid running smoothly. Adapted to the advanced hardware, these programs ran 10 times faster on the multithreaded machine. "That was the best speed ever reported. We're getting closer to being able to track the grid in real time," said Nieplocha.

In biology, another complex web is woven by genes (or their protein products) working together inside people's cells. "We have discovered genes implicated in breast cancer using a massively multithreaded algorithm on the Cray XMT," said Georgia Tech computational scientist and engineer David A. Bader. "It's like finding a needle in a haystack. The algorithm searches for genes whose removal quickly causes networks and pathways in the cell to breakdown."

The processors and computer memory in the advanced computers interact in a novel way. In traditional supercomputers, each processing chip gets a dollop of memory to use for its computations. To perform a calculation, the chip dips into the memory, does its work, then accesses the memory again for its next calculation, like an elephant dipping its trunk into a bag of peanuts and eating them one at a time. Each processor-memory unit is linked together over a network, and performance improvements come with more and faster processors and sleek network connections.

The Cray XMT multithreaded system lumps all the memory together, and the processors freely access the much larger memory pool. But like an elephant with many trunks, each processor has multiple threads: it dips into memory with one thread, and while that thread is performing the calculation at hand, another thread goes into the memory, and another.

By the time all the threads have dipped, the original thread has finished its calculation and is ready for another trip to the memory bank. A many-trunked elephant would have a distinct speed advantage plowing through a bag of peanuts over its hungrier zoo-mate, just as a multithreaded system does.

"The processors are doing useful work all the time, so the computer can be faster," said Chavarría. Each Cray XMT processor has 128 hardware threads with which to access the shared memory.

Conceptually, this advantage translates into the machines being able to handle complex, random networks of data. Mainstream machines split up the data, assigning parcels of data to individual processing units. For example, a supercomputer trying to model how a community of microbes behaves would subdivide the community spatially.

The computer would then analyze what goes on within each subdivision, but it couldn't reach across other subdivisions to find out what happened to the microbe that wandered off to the other side of its habitat. Multithreaded machines, however, can examine the whole space at once, essentially assigning each thread to a microbe.

"If all of your microbes move to the other side of the territory, it doesn't matter, because the threads still have access," said Chavarría.

Another advantage multithreaded machines have over mainsteam computers is in power consumption. Although the Cray has not yet been tested, other multithreaded machines have shown reduced energy usage compared to traditional architectures.

Source: Pacific Northwest National Laboratory / PhysOrg.com

Related stories:

AMD Releases Specification Designed to Enable Real-Time Performance Optimization for Software Applications
AMD today made available a new specification describing “Light-Weight Profiling” (LWP), a technology designed to increase the performance of software applications by providing a mechanism that allows software to more effectively leverage the benefits of multi-core processing.
'Not so fast, supercomputers,' say software programmers
The fastest of the fastest computers - supercomputers used at national research centers, research universities and major corporations - will soon gain even more performance by taking advantage of multicore computing.
Engineer Creates First Academic Playstation 3 Computing Cluster
The Sony Playstation 3, Xbox and Nintendo Wii have captivated a generation of computer gamers with bold graphics and rapid-fire animation. But these high-tech toys can do a lot more than just play games. At North Carolina State University, Dr. Frank Mueller imagined using the power of the new PS3 to create a high-powered computing environment for a fraction of the cost of the supercomputers on the market.
AMD launches Dual-Core processors for servers and workstations
AMD today introduced a broad portfolio of dual-core solutions from server and workstations for enterprise users to desktop and desktop-replacement notebooks for consumers and business users. At the second-anniversary celebration of the award-winning AMD Opteron processor, the company announced immediate availability of the Dual-Core AMD Opteron 800 Series processor for four- to eight-way servers. The 200 Series processors for two-way servers and workstations will be available in late May. The new processors deliver up to a 90 percent performance improvement for application servers over single-core AMD Opteron processors. For desktop PC users, AMD announced the new AMD Athlon 64 X2 Dual-Core processor brand that will enable true multi-tasking capabilities for richer computing experiences.
Dell to Include Intel Dual-Core Processors
Dell today announced that its workstation and consumer desktop lines will be among the first to include the enhanced performance enabled by Intel's new dual-core processor technology. Available in the United States and Canada, the new systems help customers save time and use computing resources more efficiently for design and analysis or when creating or enjoying digital content.
First PCs based on Intel Dual Core Processor Systems Appeared
Velocity Micro, a leading North American manufacturer of high performance PCs, announced recently that it has partnered with Intel to introduce a new line of systems based on Intel's new dual core processor technology. Featuring the Intel Pentium 4 Extreme Edition 840 processor in its new DCX line of computers, Velocity Micro will ship systems tuned to operate at up to 4.0GHz at introduction.
Intel Xeon Processor Family To Boost Performance, Memory and Graphics for Workstation Platforms
Platform Innovations Provide Performance and Headroom for the Most Demanding Applications

SANTA CLARA, Calif., June 28, 2004 - Intel Corporation today announced availability of a new Intel® Xeon™ processor-based platform and a host of new products and technologies for its Intel Xeon processor family that significantly boost performance, memory and graphics capabilities for workstation platforms. Workstations will benefit from rich set of new technologies that address the increasingly data-hungry systems and software applications that crave performance for a range of functions such as financial and scientific data modeling to digital filmmaking and design automation.
Intel Upgrades Software Tools to Support Mac OS X Leopard
Intel Corporation today announced an upgrade of its popular software tools suite for Mac OS X Leopard, the sixth major version of Apple's advanced operating system. The latest 10.1 version of the Intel C++ Compiler and Intel Fortran Compiler, as well as the Intel® Threading Building Blocks, Intel Math Kernel Libraries and Intel Performance Primitives, have been optimized for Apple's Leopard and Xcode 3.0 development environment launched last month.

News discussion:

Technology news