The Metanet White Paper - Workshop on Research Directions for the Next Generation Internet John T. Wroclawski Research Scientist MIT Laboratory for Computer Science 545 Technology Sq. Cambridge, MA 02139 jtw@lcs.mit.edu 617-253-7885 voice 617-253-2673 fax The Metanet John Wroclawski MIT Laboratory for Computer Science This white paper suggests that a global network architecture incorporate *regions* as first-class objects, and describes the Metanet, one such architecture. 1) Regions Today's Internet is conceptually built on the model of uniform end-to-end connectivity, with routing providing direct paths between any set of end-nodes. In reality, the global network consists of many separate regions interconnected with varying degrees of trust and cooperation. This difference between theory and practice has led to a number of ad hoc creations such as firewalls and NAT boxes. This is an unfortunate situation. The very elements that match the Internet architecture to real world administrative, security, and policy enforcement requirements, and thus make the net actually useful, are add-ons rather than part of the architecture. We argue that a new architectural component, the region, should form a central building block of the next generation network. In this model, the network is fundamentally viewed as a composed collection of regions. The basic operations of the network - data transmission, resource management and control, routing and addressing, security, management and configuration, and pricing - take place in this context. A region is a partition of the network which models an externally defined or imposed structure. The region captures the concept of an area of consistent control, state, or knowledge. There can be many sorts of regions at the same time - regions of shared trust, regions of physical proximity (the floor of a building or a community), regions of payment for service (payment zones for stratified cost structures), and administrative regions are examples. Within a region, some particular invariant is assumed to hold, and algorithms and protocols may make use of that assumption. The region structure captures requirements and limitations placed on the network by the real world. Two concepts are closely associated with regions; boundary crossing and membership. The first is tied to connectivity and routing. Even in a network of regions, the goal is end-to-end communication. Paths of communications must thus be established in a mesh of regions, which implies passing through points of connection between the regions. We call these points waypoints. Concrete examples of waypoints include firewalls, proxy servers, and accounting meters. A technical problem which must be solved to incorporate waypoints into the architecture is that of constructing routes which guide data through the necessary waypoints, and insuring, to the trust level required by the customer, that the transmitted traffic has in fact done so. As an example, the crucial issue with a firewall is to insure that it has not been bypassed. This is done with ad-hoc methods today. Architectural support for waypoint location and traffic routing allows protocols based on trusted action at region boundaries to become an integral part of the system. The second concept crucial to a region architecture is membership. Because all of the members of a region share some common property, it is frequently useful to ask whether a node is a member of some region. A powerful example is found in the relation between regions and search operations. In a global network, most search operations are bounded or scoped to limit the field of search to relevant material. Often, a region is an ideal candidate to scope a search, because the region reflects an external structuring that creates a natural constraint boundary for the search. Our key challenges are to identify and justify the core properties of the region, to recast the successful architectural principles of today's Internet to this new architecture, and to show that this architecture supports networks which are more robust, scale more gracefully, and are easier to manage and deploy than the existing Internet. 2) The Metanet We call a network which builds coherent user level semantics from a regionalized infrastructure and qualitatively heterogeneous communication technologies a Metanet. Three essential aspects of the Metanet are a routing and addressing system designed for region-based networks, end-to-end communication semantics based on logical, rather than physical, common data formats, and abstract models for QoS and congestion management; mapped to specific technologies as required. Today's networks interconnect by sharing a coherent low-level transport model - the Internet is defined by the universality of the IP protocol layer. But the uniformity of the IP layer is not a critical issue to the user constructing a distributed application. What the user cares about is more basic - how the communicating parts of his application name and locate each other, the level or quality of service he can expect to obtain, and the relationship between his application's content as it enters the network and as (perhaps transformed by format conversion or computation within the network) it leaves. In fact, data need not be carried in the same way in different parts of the network - any infrastructure which meets the user's requirements with high confidence can be used to construct a coherent application. Packets, virtual circuits, analog signals, or other modes, provided they fit into a basic service model, are equally suitable. The overall network may contain several regions, each defined by the use of a specific transmission format. For example, a single logical network might carry voice calls over a telephone infrastructure in some places and a packet infrastructure in other regions. This use of heterogeneous lower layers in different parts of the network offers powerful benefits in efficiency, deployability, adaptation to local conditions, and cost. The diversity: o Enables the use of highly-tuned special purpose physical infrastructure in regions where performance (speed, low power, weight) or economic requirements warrant. o Supports the graceful integration of legacy infrastructure into an evolving system. o Simplifies the use of any available technology in time of need or crisis. The Metanet concept raises numerous research questions. We mention two: o Network addressing, routing, and resource management systems that explicitly recognize the network as a collection of regions: - A single, global routing and addressing architecture spans different data representation regions. - The routing and addressing architecture must support assured routing across specific intermediate points in the network, as well as between end-points. o Abstraction of internetwork-layer function: - The internetworking data unit serves two functions; it is the fundamental unit of data transmission for the application, and it is the unit on which core networking functions operate. Today this is the packet. Our task is to identify -logical- data units which meet application and network requirements, and to express these core network functions - congestion control, shared resource management, and data forwarding - in terms of these logical units. - These logical functions are then mapped into the concrete behavior required by a particular technology. This allows the same abstract mechanism to control, for example, packet generation in a PC and channel allocation in a spread-spectrum radio. Our task is to develop the unifying principles underlying these mappings.