Over the past 20 years, performance gains in computing power have inspired cost effective alternatives to the traditional supercomputer. In fact, three of the ten fastest computers in the world are alternative systems. Faster and cheaper processors, faster data switching, and open source software have all contributed to new computing architectures that harness the power of supercomputers.
But this type of computing power is no longer limited to academia and those who can afford the high price of a traditional supercomputer. Corporations around the world need high performance computing to maintain their competitive advantage. A handful of quick examples include mining large databases (Fingerhut), designing microprocessors (Intel), developing new pharmaceuticals (Merck), and simulating America’s Cup yachts in different race conditions (Farr Yacht Design).
Three primary alternatives are in use today; Massively Parallel Processing, Beowulf Clusters, and Networks of Workstations.
The first alternative architecture to gain momentum was called massively parallel processing MPP. These systems differ from traditional supercomputers because they use commodity (though still high performance) scalar (common PC) processors instead of state-of-the-art vector (wicked fast) processors. Two current systems include a data mining applications at Fingerhut and a system for modeling nuclear reactions.
Intro to MPP
Sandia National Lab’s Red Storm
A Beowulf cluster simply refers to any set of common computers interconnected to perform a single, complex computational task. Generally, it is defined by three characteristics: 1) off-the-shelf PCs, 2) fast data switching connection such as fast Ethernet or fiber, and 3) an open source software such as Linux. It is one degree removed from parallel processing in the sense that MPP holds all the components in one physical box while in a Beowulf Cluster the components are individualized into each PC. The fastest cluster is ranked 3rd in the world in terms of computing power and is located at the Lawrence Livermore National Laboratory.
A Note From the Creator
Beowulf Players and Applications
Networks of Workstations (NOW) was the next logical step from Beowulf Clusters, utilizing improved networking technology, and accelerated development time to distribute computing processes to underutilized computers scattered through a network.
A key evolutionary feature from MPP is the substitution of distributed memory for server based memory or disk. By utilizing excess RAM on the network, performance is increased dramatically because distributed memory far outstrips space available on a local server. This concept of distributed storage capacity can be elevated to disk space with software that stores the same data on multiple workstations, making it immediately available to the main process even if disks are being accessed for other calculations.
Networks of Workstations are deployed within organizations like in a building or corporate intranet where all the workstations can be controlled.
UTC Uses NOW
Two additional technologies that have not yet reached the mainstream will shape computing in the coming decades. Grid computing is already beginning to develop, while quantum computing is 10 -15 years away from helping to solve corporate problems.
Grid computing further decentralizes the computing environment by allowing users to utilize computational resources that they do not own or control. In a grid, nodes on the internet can be called upon to perform tasks, though each node has a different resource manager that must approve the task. Grids are the infrastructure on which utility computing (pay per service) will be based in the future.
Grid Computing Defined
Grid Computing Luring Mainstream (Microsoft, IBM, Sun) Backers
Gartner Comments on the Future of Grid Computing
The world we see every day is based on classical physics. The state of every object can be defined precisely. For example, the lights are either on or off, those are the only two options. Inside our computers, the microprocessor can only be in one state at a time, and can therefore only perform one calculation at a time.
Classical physics describes the macroscopic world well, but its simplifying assumptions break down when we look at a very small scale. If we consider individual atoms interacting with each other, we are in the realm of quantum mechanics. The world of quantum mechanics has very different rules and requires different intuition. If we consider a quantum mechanical light, the light could be on, the light could be off, or the light could be both on and off at the same time.
Processors based on quantum effects allow parallel processing to occur on the same chip. The first significant algorithm written to utilize a quantum computer was a factoring algorithm, which has implications for today’s encryption techniques. Quantum computing is required to simulate all microscopic systems governed by the laws of quantum mechanics, one example is nanosystems.
Quantum Computing is Out There, and it Just Got Funding
IBM Test Tube Quantum Computer Makes History