Address: Peter A. Steenkiste Senior Research Scientist School of Computer Science Carnegie Mellon University 5000 Forbes Avenue Pittsburgh PA 15213 e-mail: steenkiste@cs.cmu.edu Tel: (412) 268-3261 Fax: (412) 268-5576 A Networking Research Agenda Driven by an Electronic Services Market We see the emergence of a electronic service industry that is eager to deliver a wide variety of services to end-users. Services will range from low-level "bearer" services that transport bit streams over the network infrastructure to value added services such as video conferencing, computing services, and data mining. Applications will use these electronic services to implement sophisticated end-user functionality. An example is an enhanced video conferencing application that supports cooperation among multiple parties by combining video conferencing with access to large amounts of archived data, real time data streams, and distributed computing tasks. Such an application would heavily rely on computing, storage and communication services provided by the network, and might use only minimal resources at the end-points (e.g. PCs). We envision that this electronic services market will be very dynamic and will have a hierarchical structure: users will be able to "hire" services quickly and for short-term tasks, creating a very competitive environment, and higher-level services will rely heavily on lower-level services to implement functionality. While the detailed organization of this electronic services market will be driven by market forces, the network research community has to develop technologies that can support a competitive electronic market place in an effective way. Innovation is needed in a number of areas. First, the network will have to handle traffic streams with quality of service (QOS) requirements much more diverse than those being defined today. Second, the network has to support the sharing of resources between cooperating traffic streams, i.e. besides supporting QOS at the level of individual connections, it also has support a QOS model at the application level. Third, we have to develop systematic and efficient methods for balancing the constraints and priorities of services competing for network resources. Fourth, we need dynamic mechanisms for electronic commerce. Finally, mechanisms have to be put in place so that services can be provided in a robust and secure manner. In each of the above areas, the electronic services market dictates a broad set of requirements that will have an impact on key network design issues. We briefly touch on two such problems: congestion control and quality of service. The congestion control problem deals with situations in which the demand for network capacity exceeds the available resources, i.e. multiple services are competing for a scarce resource. Networks today use a variety of mechanisms to avoid or recover from congestion. They range from relying on end-points voluntarily reducing their transmission rate in response to dropped packets (the current Internet), to more sophisticated (and complex) schemes in which switches provide explicit feedback to traffic sources that adapt their transmission rate in a standardized fashion (ATM ABR traffic). These mechanisms explicitly or implicitly try to implement a fairness model where resources (e.g. bandwidth) are distributed equally across different connections. While this generic model is appropriate if no information to distinguish between sources is available, an electronic market will provide higher level information that will make it possible and necessary to implement more sophisticated fairness models. Guidance can come from the market, e.g. two competitors have a contractual agreement on how they will share resources on a congested link, or from application requirements, e.g. "cooperating" connections have a preset policy for sharing resources. While the specific policies will evolve over time, the networking research community has to develop the mechanisms that will make it possible to implement these flexible policies. Another example of an area that will be heavily influenced by a diverse electronic services market is the area of QOS definition and implementation. The network community is currently pursuing the notion of an integrated services network that can support a number of classes of service, including both a variety of guaranteed and best effort services. The service classes specify what service (e.g. loss rate, latency, bandwidth) a specific connection can expect if it limits its traffic to a negotiated amount. In contrast, services providers will focus on optimizing a much higher-level of QOS: the QOS applies to a set of connections belonging to single application, and not to individual connections, and may involve the services provided by computing and storage nodes. These types of QOS will be as diverse as the services themselves. An example of a higher level QOS requirements is: guarantee that a new participant to a video conference can be included in x seconds. These services are likely to be implemented on top of lower level network services, although we expect that many changes will be required to the lower levels. One interesting area where there is a lot of room for innovation is that of service classes that sit between best effort service and services that provide "hard" guaranteed services. Existing markets suggest that there will be a demand for services that are looser than the existing guaranteed services, but that are more predictable than best effort. One example looser guarantee (the equivalent to "delivery within 30 days"). Another example is a service class that effectively supports network aware applications. This service class could for example make network behavior more predictable on a time scale that is appropriate for the application, or it could help in providing applications with information about network status, so they can adapt more easily. A final example is an service targeted at applications and services that have advance knowledge of their resource requirements. By passing this information on to the network, they could either get better service, or a reduced charge. New mechanisms are needed to support these types of service classes. We believe that a rich electronic services market will develop, and that it is important that the research community develops the technology needed to make this market run in an organized, efficient and secure way. Resource management plays a central role: how can network resources such as bandwidth, scheduling capacity, storage, computation, be managed in a flexible and dynamic manner. We can identify three dimensions to the problem: resource allocation in the "space" consisting of the physical network infrastructure and attached processing resources; decision making on different time scales, ranging from application startup to packet and cell scheduling; and resource allocation by different organizational entities. In all three dimensions, the mechanisms have to provide for extensive customization to application requirements.