Our research group is part of project BLANCA, a gigabit network testbed funded by the Center for National Research Initiatives. It involves research teams from AT&T Bell Laboratories at Murray Hill, NCSA, Sandia National Laboratories, Lawrence Livermore National Laboratories, Lawrence Berkeley Laboratories, University of Illinois at Urbana-Champaign, University of California at Berkeley, and University of Wisconsin. The sites are interconnected by 45 megabits/sec links and selected trunks are interconnected by 622 megabits/sec links including the connection from the University of Illinois to the University of Wisconsin. The gigabit testbeds are being used to show the benefits a national gigabit network would bring to the scientific and engineering community and enable new technology like distributed supercomputing and data visualization. Our part in the project has three components: (1) design of the switch controller software for the XUNET 600 megabit/sec switches that are used in the network; (2) design of HIPPI to XUNET interfaces to connect various supercomputers to the network; and (3) conducting high-performance distributed computing experiments using the BLANCA testbed to explore the benefits of the larger bandwidth. Shared memory communications are supported over BLANCA by an implementation of distributed shared virtual memory developed at the University of Illinois.

A local ATM network, interconnected with XUNET-based BLANCA gigabit network, is being used as the backbone for digital video and multimedia studies. Research includes multicasting, compression, and protocols. This project is also investigating the interoperability of different ATM switches and networks, both local area and long distance. The research includes the study of signaling protocols, switch communication schemes, call set-up, and the design of switch controllers.

The infrastructure evolves through the five-year project plan from simple subnetworks for parallel processing and storage, ATM studies, and collaborative multimedia computing to a hierarchy of ATM networks with different communication bandwidths and a variety of digital media services. Matching funds from the campus will allow us to explore the extensibility and scalability of the infrastruc- ture for activities like remote collaboration, data archiving and prefetching, immediate and time-shifted learning, multimedia-based education, and interactive digital video communication.

This project is studying new high-bandwidth networking technologies based on GaAs integrated electronic and optical components. Various key network components are being built to evaluate the approach. The Pulsar switch design is built from simple components with the aim of maximizing performance by minimizing complexity. At the heart of the switch design is a shift register ring that can support the switching of packets at bandwidths in excess of 128 gigabits/sec. A prototype Pulsar switch has been built and evaluated. A faster gigabit/sec implementation of the Pulsar switch is in the program for summer 1995.

The goal of this project is to develop low-cost, high-speed networks for scalable server applications where ATM may be too expensive. To achieve high performance, these networks must deliver controllable delay and guaranteed bandwidth for continuous media traffic as well as high availability. Traffic models that must be supported include CBR, ABR, and VBR.

Low-latency and high-throughput communication are performance-critical in both parallel computers and workstation clusters (scalable servers). Our objective is to develop software and hardware architectures for scalable servers in the national information infrastructure. The architectures must achieve communication performance which scales in bandwidth, packet rates, connection rates, as well as supporting continuous media data types with real-time requirements (video, audio, etc.). We have developed extremely high-performance implementations of TCP/IP, UDP, MPI, PVM, etc., and new protocols are being considered. Network interface architecture issues include supporting protection, multitasking, distributed scheduling, and distributed process management.

In view of the disparate nature of the data traffic through a typical switch, some transmissions must be carried out under given timing constraints. Detailed timing constraint satisfaction requires information about the data content of the cells (and not just network-level parameters). Guaranteeing a certain ATM cell throughput does not ensure constraint satisfaction on underlying data transfer. More importantly, the throughput constraints are on the cell contents (and sometimes relative to the block-level parameters). The goal of this project is to capture detailed timing constraints that are applied on the source data and to develop constraint analysis techniques for use in a switch scheduling algorithm. This ensures that relative rate constraints are satisfied.

We are investigating the properties of various bus connection networks (viewed as generalized incidence systems, such as projective planes). The mathematical properties of such networks suggest some intriguing new techniques for performing both computation and communication. Our investigation centers on four topics: (1) discovery of simple, expandable, and inexpensive interconnection strategies which can be applied to very large networks, having possible thousands of nodes; (2) specification of simple, adaptable message routing schemes which can be implemented on those networks; (3) network survivability under partial degradation; and (4) discovery of practical applications of such networks.

In this project we investigate algorithms in services and protocols for communication systems to support guaranteed transmission of audio/video streams according to quality of service (QoS) specification. We study admission control for resource availability, negotiation protocols for balancing QoS requirements among distributed end-points, and resource reservation allocation mechanisms. Furthermore, we investigate QoS monitoring, adaptation, and renegotiation services to capture the network's dynamic behavior.