, University of California, Berkeley
, Com21, Inc.
, Hewlett-Packard Laboratories
Pages: pp. 10-11
Welcome to the special issue of IEEE Micro dedicated to the best of the presentations at the Hot Interconnects III Symposium held at Stanford University in August of 1995. The annual symposia provide a forum for the latest technology developments in interconnections from optical to copper to wireless, switches to network interfaces to software protocols, and supercomputer backplanes to local area and wide area networks.
At Hot Interconnects, professionals from academia and industry exchange ideas and information about issues that lie on the boundary between the disciplines of computer technology and communications. With the six articles in this issue, we convey some of the environment of our symposium. Indeed, these articles illustrate the breadth of research and development in interconnect technology.
The popular media has focused much attention on the interconnection of computers in the wide area, through the Internet. Although issues in the wide area were important at Hot Interconnects III, some of the most popular sessions and papers concerned the interconnection of computers in the Networks-of-Workstations (NOW) models and how to evaluate the performance of such architectures. Interaction between engineers and computer scientists working at each of these levels was quite lively. Although the relative importance of metrics like latency and throughput is quite different across different interconnection distances and applications, the basic issues involved are fundamental.
Presentations are selected for inclusion in this issue of IEEE Micro by vote of symposium attendees, depending on whether the presenter can produce a full article that is not on track for publication elsewhere. The interest in more tightly coupled networks of computers is reflected in the articles selected for this issue. The first four fall more in the "moderate distance" interconnect camp, and the latter two discuss interconnections for Internet access. Another popular presentation on infrared wireless networks, "Performance Evaluation of Experimental 50-Mb/s Diffuse Infrared Wireless Link Using On-off Keying with Decision-Feedback Equalization," by Gene W. Marsh and Joseph M. Kahn, will appear in IEEE Transactions on Communications.
Richard Gillett describes a network interface card, the PCI (peripheral component interconnect) bus, that automatically reflects memory writes over the network to a remote machine. He shows that this approach offers both low latency and high bandwidth, demonstrating tangible benefits for end-user applications.
David Engebretsen, Daniel Kuchta, Richard Booth, John Crow, and Wayne Nation describe a low-cost, 2-Gbyte/s, parallel optical link for moderate-distance (100 meter) processor-to-processor interconnection. They have integrated their link technology with the Scalable Coherent Interface (SCI) standard to demonstrate its usefulness for actual computing applications.
Supercomputers often stress the envelope of interconnect technology. Steve Scott describes Cray Research's next-generation system interconnect, a 800-Mbyte/s full-duplex ring. Scott enhanced his SCI-based system to address the need for reliability and "hot swaps" in a distributed supercomputer interconnect.
Many technologists start from the perspective of "if you build it, they will come." Equally important is research in measuring the performance of actual systems to evaluate the impact of design trade-offs. Two articles address this need. David Culler, Lok Liu, Richard Martin, and Chad Yoshikawa devise a set of microbenchmarks for use in teasing apart CPU overhead, wire and switch latency, and bandwidth for a variety of network interfaces. One conclusion is that small-message round-trip times on a general-purpose LAN can be within a factor of two of that provided by a specially designed multiprocessor interconnect.
Wireless computing is another area in which the promises of technology need experiential validation. Stuart Cheshire and Mary Baker measured and evaluated a commercial packet radio system in the San Francisco Bay area as a mobile Internet service.
Looking ahead, Sandy Fraser of AT&T discusses the potential basis for integrating the Internet and standard telephone service. He argues that the wide area network switches for this integrated service are likely to use lots of memory to simplify flow control. Fraser shows that the cost of the DRAM to buffer all in-flight data is at least three orders of magnitude less than the cost of the wide area transport itself.
Many dedicated volunteers on the organizing and program committees helped develop Hot Interconnects. We sincerely thank all the people who helped make the conference successful. We are also grateful to Stephen Diamond for the opportunity to publish this set of articles in IEEE Micro.
Hot Interconnects IV meets at Stanford University on August 15-17, 1996. Contact 1996 general chair Qiang Li at firstname.lastname@example.org for information about the symposium. Contact program cochairs Kai Li (email@example.com) or Chuck Thacker (firstname.lastname@example.org) for information about presentation submittals, which are due March 18, 1996.
Indicate your interest in this article by circling the appropriate number on the Reader Service Card.Low 150 Medium 151 High 152