The AURORA gigabit testbed

David D. Clark, Bruce S. Davie, David J. Farber, Inder S. Gopal, Bharath K. Kadaba, W. David Sincoskie, Jonathan M. Smith, David L. Tennenhouse
1993 Computer networks and ISDN systems  
AURORA is one of five U.S. networking testbeds charged with exploring applications of, and technologies necessary for, networks operating at gigabit per second or higher bandwidths. The emphasis of the AURORA testbed, distinct from the other four testbeds, BLANCA, CASA, NECTAR and VISTANET, is research into the supporting technologies for gigabit networking. Like the other testbeds, AURORA itself is an experiment in collaboration, where government initiative (in the form of the Corporation for
more » ... ational Research Initiatives, which is funded by DARPA and the National Science Foundation) has spurred interaction among preexisting centers of excellence in industry, academia, and government. AURORA has been charged with research into networking technologies that will underpin future high-speed networks. This paper provides an overview of the goals and methodologies employed in AURORA, and points to some preliminary results from our first year of research, ranging from analytic results to experimental prototype hardware. This paper enunciates our targets, which include new software architectures, network abstractions, and hardware technologies, as well as applications for our work. Several approaches have been proposed t o achieve the next generation of network, based on different information transfer paradigms. AURORA will explore two significant options for network architecture and the interworking between them. The Asynchronous Transfer Mode (ATM) based on the transfer of small, fixed-size data cells, is the broadband methodology currently favored within the telecommunications industry [20]. Packet Transfer Mode (PTM) is the term used in this document to describe packet transport methodologies that permit a mixture of different packet sizes within the networkl.lt is the method preferred by segments of the data communications industry. Each approach has its advantages, and they will coexist in the national network of tomorrow. This project will enhance and deploy experimental switch prototypes tailored t o each of the transfer modes -Bellcore's ATM-based Sunshine switch and IBM's PTM-based plaNET switch. Since switches are only one aspect of network technology, the project will prototype additional components, both hardware and software, to support the associated transmission, signaling, interface, operations and management functions. Transfer mode independent issues, including higher level transport protocols, are being explored, and the architecture of host interfaces is under investigation. The interworking issues between the PTM and ATM environments will also be studied. A result of this experiment will be hands-on experience with the use of these two transfer modes, a characterization of the domain of utility for each of them, and an increased understanding of the problems of internetworking at gigabit speeds. Distributed System/Application Interface Paradigms An important part of network architecture is packaging the network service and presenting it to the application builder in a way that simplifies the application design without restricting or unduly complicating the operation of the network. This packaging, often called "abstraction", is a basic aspect of computer science. The most popular abstractions for today's networks are the reliable byte stream and the remote procedure call (RPC). Both of these seem t o provide convenient and natural interfaces to applications, but both are limited in the functions they can deliver. The byte stream, because it insists on reliable delivery, cannot also control latency of delivery. Remote procedure calls, because they represent serialized rather than parallel communication across a network, degrade directly with increasing network latency. For high-speed networks, RPC makes poor use of the available bandwidth. An alternative network abstraction is one in which the network is modeled as shared virtual memory. That is, the application makes use of the network by reading and writing parts of its address space which are replicated at the communicating sites using the network. This abstraction is a very basic one t o computer science, but potentially suffers from the same latency problems as RPC. However, techniques from existing virtual memory management implementations and file systems, namely read-ahead and caching, can be adapted to the network context. This approach stresses network transparency and assumes that the software supporting the application interface can deduce the proper action (e.g., read-ahead) from the past behavior of the application. In contrast, alternative approaches that are less transparent and require some explicit characterization of application service requirements are also being explored. This approach might serve a broader set of applications than an implicit scheme such as shared virtual memory. However, 'It should be noted that while ATM transports data through the network in fixed-sized packets, applications may still communicate using variable length packets. Conversion between ATM cells and variable length packets is handled in a host interface, described in Section 4.2
doi:10.1016/0169-7552(93)90056-a fatcat:6mc6vxelhnhphc4nsnkliejany