Interactive retrieval in information spaces distributed across a wide-area network.

摘要

The potential to provide interactive data manipulation across high-speed nationwide networks is stimulating development of new database technology. An information space is a data model that can support rapid browsing of large amounts of information contained in a digital library physically distributed across many disparate sources. This dissertation discusses supporting interactive retrieval of objects inside an information space across the nationwide scientific network. Implementing such interactive retrieval requires designing caching policies that enable fetching requested objects into a local user workstation from a remote file server with sufficiently short response time to support effective browsing interaction. An adequate caching policy should utilize properties of user perception and data representation within an information space. This dissertation describes a series of new techniques for caching objects within an information space and gives measurements of their performance across the NSFNET. These policies take advantage of special features of interactive retrieval within information spaces, such as initially fetching only the subset of requested objects that will be immediately displayed and prefetching additional objects during idle time when the user is considering which command to issue next. A prototype built by the author, the Telesophy System, supports interactive retrieval for information spaces across local-area networks and serves as a basis for identification of special features. To consider additional needs for efficient implementation across wide-area networks, the significant parameters and policies in implementing caching are systematically identified. Specific values of these caching parameters are used to evaluate the performance of a range of caching policies under a variety of interactions relevant to browsing information spaces. Finally, an incremental caching policy is proposed, which combines many techniques taking advantage of special features of interacting with information spaces. Measurements of the performance of this policy under a variety of conditions demonstrate that interactive retrieval is possible across wide-area networks and that appropriate optimization of the caching policy can produce performance comparable to that across local-area networks.