Disk system architectures for high performance computing

Article Abstract:

Using redundant arrays of inexpensive disks (RAIDs) is a cost-effective approach to the high input/output (I/O) bandwidth between disk memory and processor required by high-performance computer architectures. Increasing speed of successive generations of computers has generally not been matched by advances in I/O performance. An overview of the basic technology of magnetic disk drives and I/O controllers and a review of proprietary disk subsystem design approaches by IBM, Cray and Digital Equipment Corp illustrates the nature of the problem and some responses to it. Strategies for improving the I/O performance of conventional disk systems include fixed head disks, parallel transfer disks, increasing disk density, solid-state disks, disk caches, disk scheduling and disk arrays. The design, functional levels and I/O control of RAID disk arrays are discussed.

Computer peripheral equipment, not elsewhere classified, Computer storage devices, Disk drives, Performance improvement (Computers), Trends, Case Study, Peripheral Controllers, Performance Improvement, I/O Management, I/O management (Computers)

---

Parallel architectures for vision

Article Abstract:

Processor connection autonomy in a fine-grain SIMD (single instruction, multiple data stream) massively parallel computer offers the most flexible and efficient system architecture for vision processing applications. The principle of processor autonomy constitutes the core of a new model for parallel processing architectures targeted at vision computations. The major types of vision computations are image processing, transformations, feature extraction and combination, geometric and graph computations. SIMD is found to be the most applicable of the three main classes of parallel architectures: SIMD, MIMD and MISD. A comparison of the three types of processor autonomy - operation, addressing, and connection - reveals that the latter offers the most benefits in the SIMD parallel computer vision system.

Computer vision, Processor architectures, Cost benefit analysis, Parallel processing, Applications, Processor Architecture, Machine Vision, MIMD, Massive Parallelism, SIMD, MISD

---

Similar abstracts:

• Abstracts: Array architectures for iterative algorithms. Advanced parallel processing with supercomputer architectures. Task-Oriented Architectures
• Abstracts: MUPSI: a multiprocessor for signal processor. Multiprocessor computing for images. Integrated optic approach for advanced semiconductor lasers
• Abstracts: Synchronous data flow. Performance of coherent optical receivers. The synchronous data flow programming language LUSTRE
• Abstracts: Array architectures for iterative algorithms. On the resolution and contrast of microwave images using a complete first-order iterative algorithm
• Abstracts: Real-time simulation for space station. Shuttle orbit flight control design lessons: direction for space station

This website is not affiliated with document authors or copyright owners. This page is provided for informational purposes only. Unintentional errors are possible.