The Era Of Supercomputing

The earliest modern computers were room-sized monsters that were programmed by paper tape and punch cards. Computing in the 1960's and 1970's was dominated by mainframes -- large machines that could support thousands of simultaneous users accessing the system through "dumb" terminals.

LSI (Large-Scale Integration) and VLSI (Very-Large-Scale Integration) technology developments in the 1970's and 80's meant that thousands of components could be integrated on a single hardware chip, thus leading to the development of the microprocessor. Additionally, the development of the vector CPU architecture meant that large amounts of mathematical data could be handled much faster than other types of processors, though vector CPUs could be slowed down when it came to complex instructions. The Cray supercomputer models based on vector technology introduced many innovations in the late 1970's and 1980's and were the fastest computers of their time.

The number of processors that can be clustered together in a traditional vector-processor supercomputer, however, is often limited by the ability to access the large amount of shared memory. "Massively parallel" systems were designed to overcome this roadblock by directly connecting off-the-shelf processors with local memories to each other by a network, thus enabling a great deal of scalability generally limited only by costs.

Because many problems carried out by supercomputers can easily be split up into smaller parts to be worked on simultaneously (parallelization), traditional supercomputers can often be replaced by parallel-processing "clusters", which are individual machines that are programmed to act as one computer. Beowulf is perhaps the most well-known type of parallel processing cluster today. Donald Becker and Thomas Sterling designed the first Beowulf prototype in 1994 for NASA. It consisted of 16 486-DX4 processors connected by channel-bonded Ethernet. The next Beowulf clusters were built around 16 Pentium Pro (P6) 200-MHz processors connected by Fast Ethernet adapters and switches.

Thomas Sterling has explained that the name "Beowulf" was originally intended only to designate the project and not the system itself. "Beowulf" was an allusion to the Scandinavian folk tale of the hero who killed the monster Grendel, a similar story to that of "David and Goliath", where the underdog triumphs against all odds.

Today, Beowulf parallel clusters are used to solve problems that require trillions of computations be performed quickly and efficiently. Modern science increasingly relies upon the ability to run computational simulations which are too complex, time-consuming or costly to study or model in any other way. The processing power of Beowulf clusters is applied to various complex computation problems such as mapping the human genome, weather forecasting, cryptanalysis, fractal simulation, or simulating nuclear explosions. The availability of powerful microprocessors and high-speed networks as commodity off-the-shelf (COTS) components, as well as new software components available to support high performance applications, has helped fuel the Beowulf revolution.

Into The Future...

Simulating a single heartbeat in a 3-dimensional model of the heart can take several days of processing, even on today's high performance computing systems running billions of calculations every second. As technology continues to meet the challenges of improving processor clock rates, disk space, networking speed and memory performance, supercomputing speeds will advance into the realm of Teraflops and eventually even Petaflops (1,000 trillion calculations per second).

The coming years will also bring an increase in pervasive computing (or ubiquitous computing) -- the presence of embedded technology and connectivity into everything from cars and homes, to clothing and household appliances. Many also see a move toward "metacomputing", the collection of many single computers acting as a single system. The grid metacomputing model aims to solve problems too intensive for any single supercomputer by harnessing the power of distributed individual systems over high-speed networks.

The use of supercomputing will likely expand into new areas where it might be applied in various ways. In the future, supercomputers may be increasingly used to perform functions such as analyzing foot or vehicle traffic flow, managing business services, or they may be found in the medical industry, where supercomputers could interact with large databases to improve patient treatment and diagnosis.