Supporting the Information Mesh
Karen R. Sollins · 1992
As the networked world expands, it holds the promise of a new vision of human and computer connectivity. We envision a new model for the role of the network in a distributed computing environment, as the basis or core of the Information Mesh. Our early model of the network was a reliable stream of bits, facilitating such network activities as file transfer, remote login, and electronic mail. As networks became more prevalent, we incorporated them into our models of programming, as in the various forms of remote procedure call. In both of these cases, the major impact of networking on the operating system was real time constraints on the operating system; not only were there real time constraints based on application level tasks, but also the network itself presented a further real time constraint on the operating system. As high-speed workstations and networks become more prevalent, we find ourselves able to provide a new model of the network, that will necessarily require abstractions in the operating system. This new model is one based on information that is long-lived and widely distributed; we envision upwards of 1014 objects or nodes with lifetimes of up to 100 years.. Furthermore, the model incorporates the concept of relating or linking nodes of information to form the information mesh. It is this notion of links that must be understood at the operating system level, in order to participate in this large, interconnected mesh of information. The model assumes that interesting information can be distributed around the worlci, and survives outside any particular application, application toolkit, or programming language runtime system. In addition, for the information to be interesting, it will be extremely long-lived. It may outlive ownership, storage media, file systems, and other services for locating the information, each to be replaced by successors. Thus, the names that occur in links must outlive ownership, storage facilities, and underlying organizations, yet be generally resolvable at any time.' Thus the name for a node at the other end of a link must contain no semantics that will restrict the size, mobility, or longevity of the node.