Howard P. Katseff

Incomplete hypercubes

Since a k-dimensional hypercube has 2/sup k/ vertices, these systems are restricted to having exactly 2/sup k/ computing nodes. Because system sizes must be a power of two, there are large gaps in the sizes of systems that can be built with hypercubes. Routing and broadcast algorithms are presented for hypercubes that are missing certain of their nodes, called incomplete hypercubes. Unlike hypercubes, incomplete hypercubes can be used to interconnect systems with any number of processors. The routing and broadcast algorithms for incomplete hypercubes are shown also to be simple and deadlock-free. >

A 54 algorithm for two-dimensional packing

This paper proposes a new algorithm for a two-dimensional packing problem first studied by Baker, Coffman, and Rivest (SIAM J. Comput. 9, No. 4(1980), 846–855). In their model, a finite list of rectangles is to be packed into a rectangular bin of finite width but infinite height. The model has applications to certain scheduling and stock-cutting problems. Since the problem of finding an optimal packing is NP-hard, previous work has been directed at finding polynomial approximation algorithms for the problem, i.e., algorithms which come within a constant times the height used by an optimal packing. For the algorithm proposed in this paper, the ratio of the height obtained by the algorithm and the height used by an optimal packing is asymptotically bounded by 54. This bound is an improvement over the bound of 43 achieved by the best previous algorithm.

Lower bounds for on-line two-dimensional packing algorithms

SummaryMany problems, such as cutting stock problems and the scheduling of tasks with a shared resource, can be viewed as two-dimensional bin packing problems. Using the two-dimensional packing model of Baker, Coffman, and Rivest, a finite list L of rectangles is to be packed into a rectangular bin of finite width but infinite height, so as to minimize the total height used. An algorithm which packs the list in the order given without looking ahead or moving pieces already packed is called an on-line algorithm. Since the problem of finding an optimal packing is NP-hard, previous work has been directed at finding approximation algorithms. Most of the approximation algorithms which have been studied are on-line except that they require the list to have been previously sorted by height or width. This paper examines lower bounds for the worst-case performance of on-line algorithms for both non-preordered lists and for lists preordered by increasing or decreasing height or width.

Predictive prefetch in the Nemesis multimedia information service

Critical to the success of future multimedia services is the ability to provide fast access to stored information via communications networks. In the Nemesis project, we focus on application control protocols for delivering stored multimedia to a user. We are exploring adaptive rate control schemes that make use of predictive prefetch of information from remote storage servers as a strategy for coping with short-term network congestion. For users with low-end equipment or slow network connections, these protocols provide a degraded but still useful service. We also provide “better than being there” features such as variable playback rates with intelligible audio, and synchronization and linkage of multiple media. Our first prototype gives access to a multimedia database containing talks given at AT&T, with hypermedia links between video, audio and accompanying documents and viewgraphs. Nemesis provides for on-line archival storage and remote viewing of these presentations by members of AT&T R&D community throughout the world. It is integrated with the NCSA Mosaic navigation system as well as the AT&T LINUS information retrieval system. Our goal is to drive the technology and infrastructure to allow communities and large institutions access to multimedia information anytime and anywhere.

A distributed computing environment for robotics

Meglos is a multiprocessor operating system that is ideal for real-time applications. It provides the user with a distributed programming environment for developing, running, and debugging application programs. Meglos allows a single application to span several processors. Further, unrelated programs may run independently of each other on the same processor. A simple, yet powerful, mechanism for communications between programs is provided. With the current hardware configuration, up to twelve Motorola 68000 Multibus computer systems and a DEC VAX host running the UNIX operating system can be connected via the S/NET interconnect. Meglos exhibits communications latencies of 750 μsec, sufficiently small for most real-time applications.

Using data partitioning to implement a parallel assembler

A technique for implementing algorithms on a multiprocessor computer system is data partitioning, in which input data for a problem is partitioned among many processors that cooperate to solve the problem. This paper demonstrates that data partitioning is a good method for implementing an assembler on a message-passing multiprocessor system: it yields a speedup exceeding a factor of 6 with eight processors. We compare several alternative methods for distributing program text and sharing global information among the processors executing the assembler: operations that are important for a variety of applications implemented with data partitioning.

Communications in Meglos

Meglos provides a user‐level, message‐based programming environment for a system of interconnected processors. It allows the simultaneous execution of a wide range of applications including real‐time control and large‐scale computations requiring the co‐operative work of many processors. The communications mechanisms in Meglos are designed to be easy to use and to provide high performance at the applications level. The communications primitives are based on symmetric, error‐resilient channels that are independently flow controlled. Two‐ended channels between pairs of co‐operating processes provide communications with low latency and high throughput. Multicast channels are used to send messages efficiently among a large group of processors in a single operation.

The Liaison network multimedia workstation

A description is given of a workstation architecture, Liaison, that solves the high bandwidth problem for real-time video and allows the exploration of these communications issues. By distributing the workstations intelligence over a high-performance network, both the computation and communication load are shed to other processors on the network. The Liaison display performs only the low-level functions of acquiring and displaying images, whereas the remainder of the workstations functionality is performed by a pool of distributed processors. Initial experiences with this architecture using a prototype monochrome workstation have provided insight into the issues surrounding the control and processing of communications-intensive applications.<<ETX>>

The evolution of HPC/VORX

HPC/VORX is a computing system that provides closely coupled computing between large numbers of processors. It also supports the connection of many host workstations which may be geographically distributed within the area of a large building and allows a single applications to span many processors and many workstations. We relate some of the lessons that were learned while building and using HPC/VORX and in the transition to HPC/VORX from a smaller, less capable system. The problems that we encountered included difficulties in scaling resource managers and human interfaces to large numbers of processors, the design of communications primitives and protocols, and the implementation of programming abstractions.

Communications-intensive workstations

The introduction of gigabit local area networks and the resultant proliferation of multimedia applications will require fundamental changes in the design of computer workstations. The workstation is implemented as a distributed application on a local area network in order to focus on communications issues. A single processor is dedicated to the low-level functions of acquiring image and audio streams from the network and sending keyboard and mouse information to the network. The display processor receives only bitmap images of data to be displayed. Higher level functionality, like that provided by the X-window system, is handled by other processors in the network. To demonstrate the feasibility of this architecture, a prototype monochrome (bilevel black and white) workstation was built that is able to simultaneously display several windows with 30 frame/s video, each arriving from a different processor via a local area interconnect. The feasibility of interleaving real-time video images across multiple disks on different processors to provide sufficient throughput for full-motion video is demonstrated. A simple method provides video and sound which are synchronized with each other and which flow smoothly.<<ETX>>