EXTENDING UNIX FOR SCALABLE COMPUTING

Because it retreives all instructions and data from a single memory, the highly prevalent von Neumann computer architecture has a fundamental speed limit. The speed of light limits the rate data can move from a single memory, thereby limiting the speed of the computer. Attempts to push the physical limits in supercomputers reduce their efficiency; in fact, a personal computer provides more performance for the dollar than a supercomputer. Nonetheless, continued progress in science and engineering requires faster supercomputers. Consequently, researchers have long investigated non-von Neumann computer architectures. The scalable multicomputer architecture, which uses many microprocessors together to solve a single problem, may be a solution. In fact, scalable parallel computers that run at tera (10(12)) floating-point operations per second are now under construction. While Tflops processor technology is well known, the scalable operating and I/O system technology necessary for those speeds in not. The authors describe how Unix can be extended to scalable computing, permitting Tflops speeds and offering parallel computing to users unfamiliar with parallel programming. They designed this technology into the system software of the Ncube-2, the predecessor to Ncube's announced Tflops parallel computer. The authors describe the system specifically and provide some performance numbers.