High-Level Infrastructure for Applications in the Internet 2 Environment Clifford Lynch Director, Library Automation University of California, Office of the President Clifford.Lynch@ucop.edu Much of the discussion of Internet 2 as a development testbed for next generation network applications has focused on the relatively obvious class of efforts that can be explored initially by simply provisioning the network with a great deal of bandwidth and then later, as the network becomes more congested, with quality of service (reservation) support to manage that bandwidth. Telepresence, distance education, and other high bandwidth, near real-time transmission are all characterized by a requirement for bandwidth intensive, low latency and low loss network transport services. I believe that Internet 2 represents an opportunity to do much more than this: to build a new and more capable infrastructure to support the design and development of new classes of applications from the bottom up, and in doing so to cement new collaborations with computer scientists and computer-communications networking researchers. Multicast services represent an architectural foundation for a very large and poorly explored new class of distributed system network applications. These include distributed but federated information caches and databases, origin-driven information distribution models (including all kinds of news and telemetry, in the broadest sense), and sensor array systems, as well as more routine multipoint audio and video conferencing. Multicast services are also an essential underpinning for very large scale simulation, gaming, and collaboration applications; for example, the Internet Engineering Task Force curently has a working group exploring Large Scale Multicast Applications which is thinking in terms of supporting tens, or even hundreds of thousands of participants in multicast interactions which they term large scale multiplayer virtual worlds. Multicast services need not run at extremely high bandwidth to be useful; indeed many interesting applications may only require a few kilobytes or tens of kilobytes per second, but cannot scale past a small number of participants without multicast support; further, given current technology handling even very modest data rates with very large, dynamic groups of participants is still a research problem.. The current commodity internet cannot support these applications; the problem is not simply that sufficient bandwidth isn't always available, but that the routing support for multicast "channels" is only available on a very limited basis. Further, the need goes beyond routing of multicast data flows and also encompasses participation management (joining and leaving groups), which is an interesting "middleware" function. I believe that one valuable initiative that might be undertaken as part of the planning for Internet 2 would be the identification and supply of a series of "interesting" content sources with moderate bandwidth requirements that might be available as part of the basic Internet 2 service: these might include sensor data (such as weather, traffic, or seismic sensors), commercial news services, and financial marketplace telemetry. These could form the basis for a range of new applications. A particularly fruitful area to explore might be how multicast services could expand current thinking about digital library applications and their interactions with their user communities. There is a great deal that we don't understand well about distributed systems based on multicast models, including security and authentication as it relates to the management of multicast channels, the interactions between multicasting and database and file replication schemes, and how to characterize latency properties of multicast channels. The interaction between quality of service (an important focus of the Internet 2 effort in the point-to-point model) and multicast channels is very poorly understood. Indeed, even the basic technology of "reliable" multicast protocols has been an active research area, with a number of fundamentally different models having been proposed. These are currently active areas of basic research in computer science and networking which could be advanced by the availability of a large-scale testbed. An excellent survey of some of the current state of the art surrounding multicast can be found in the April 1997 issue of the IEEE Journal on Selected Areas in Communications -- Network Support for Multipoint Communication. The vast majority of today's production applications and applications-level protocols are based on the assumption of the ubiquitous availability of reliable point-to-point transport services as represented by the transmission protocol TCP. Establishing the similar ubiquitous availability of an appropriate multicast protocol or protocols as one of the engineering goals of Internet 2 will offer us the opportunity to explore an extensive range of new applications designs that employ multicast services as a fundamental component. Pursuing this approach, which focuses on extended routing functionality and host protocol support (including the relevant middleware) rather than simply increasing the bandwidth that is available to carry traditional point-to-point protocol based services offers a path which can permit Internet 2 to both serve as a testbed for migrating a great deal of current research into practice and to allow us to begin to explore new distributed systems architectures based on a richer set of basic network services. In my view, it would be a great mistake to limit our view of Internet 2 to an environment which allows us only to explore opportunities offered by current protocol models supported by greatly increased bandwidth. The systematic integration of multicast support is one way to reach beyond this narrow perspective. Attachment converted: HardDrive:CRA position paper (W6BN/MSWD) (00013724)