Beyond Browsers: Operating Systems Support for a Program-Enabled Web

People Working Directory

The beyond browsers project seeks to develop protocols for allowing programs, not just browsers, to make use of the valuable data becoming available on-line in geographically distributed repositories. To allow users to make better use of this data, we would like to support program-driven applications such as data-mining, collaborative computing for accessing and modifying geographically distributed data, and agent-based applications that move computations to data rather than the reverse. Doing so requires new mechanisms and policies to improve performance and to provide stronger semantic guarantees for reasoning about distributed data. We are currently working in four areas: with the goal of understanding what approaches to each of these problems is appropriate in the coming environment. We will demonstrate this toolkit with an example application --- Rent-A-Server --- that uses these facilities to allow overloaded web servers to "rent" spare capacity from underutilized ones to improve scalability, locality, and availability.



Recent work on: Active Naming

The shift in the use of the Internet from a distribution mechanism for static documents to a medium supporting nearly arbitrary services makes it increasingly important to provide mechanisms for locating and retrieving wide-area resources in an efficient and fault tolerant manner. Efficiency in this case has multiple dimensions: (i) data should be delivered to clients as quickly as possible to minimize latency supporting interactive interaction with Internet services, (ii) consumed wide-area bandwidth should be minimized to protect the shared underlying medium, and (iii) any caching or replication scheme utilized to minimize latency and bandwidth should guarantee application-specified levels of cache consistency to ensure that clients do not quickly and cheaply access incorrect information. To this end, we are designing, building, and experimenting with Active Names, a naming infrastructure that combines the process of locating and retrieving a named resource through the use of location-independent and application-specific programs. The following paper describes our prototype and application experiences in more detail:
  • Overview A more detailed description of our motivation, design and approach
    Applications  A description of Active Name applications
    Whitepaper (ps) (pdf) Amin Vahdat, Michael Dahlin, Thomas Anderson, and Amit Aggarwal, "Active Names: Flexible Location and Transport of Wide-Area Resources." Technical Report, February 1998.

    The beyond browsers project is supported by several government and industrial sponsors

    Mike Dahlin