Digital Libraries

Abstract: Wide-area systems for information storage and retrieval are rapidly gaining in popularity. Exam- ples include FTP (File Transfer Protocol), Gopher, and World Wide Web (WWW) archives of many types of information. Each system provides a means of naming a file or data object so that others may retrieve the information. Unfortunately, this name depends on the network address of the server and on the filename within that machine. Some flexibility is gained by using aliases for these names, but problems persist. Additionally, the use of aliases does not handle the replication of files at several sites, or the movement of a file from one site to another. The result is that these names frequently become invalid. A name that is good today may not work in a week or a month. If a name for a useful collection is passed on to others, it may not work by the time they try to use it. For these reasons, a better approach to naming is needed. In this paper we present a prototype distributed service for archiving files. Each file is given a permanent name, which can be used to retrieve the file from any site of the archive. If the con- tacted site does not have the data, it returns the current location, which is accessed automatically by the client. This allows files to be moved and replicated within the archive without invalidating any of the names. In a hypermedia environment such as the WWW, this means that the links will remain valid, avoiding the need to change each document that references this data. We have developed this system to demonstrate that this approach is feasible. We describe both a standalone service that is accessed via a command-line client, and a gateway to the service from the World Wide Web. The gateway allows the archive to be accessed from NCSA Mosaic, or any client that can use the Hypertext Transfer Protocol (HTTP). The service is assumed to be global in scope, so issues of scalability are critical. This influenced our design, and led us to experiment with caching of names, and automatic replication of fre- quently accessed files. We have, however, omitted the implementation of a distributed name resolver, using a centralized one instead. A distributed implementation would be required for an actual service, and we describe how this might be built. The task of locating an archive site is also discussed.