A Scalable File Replication Scheme for the World Wide Web

Keith Moore
(Corresponding Author)
University of Tennessee
107 Ayres Hall
Knoxville, TN 37996-1301
Office: 615-974-3126, FAX: 615-974-8296
http://www.cs.utk.edu/~moore

moore@cs.utk.edu

Jason Cox

cox@cs.utk.edu

Stan Green

sgreen@cs.utk.edu

Reed Wade

wade@cs.utk.edu

Abstract

The World Wide Web has reached the point where many popular file servers are overloaded, resulting in degradation or unavailability of service. Users are also often separated over a small-bandwidth pipe from a file server they need to access, resulting in poor response time. The Bulk File Distribution (BFD) system described in this paper aims to alleviate these problems by providing mechanisms for registering and looking up alternative locations for replicated files. The system also includes authentication and integrity checking mechanisms. Unlike other proposed name resolution systems, BFD provides a straightforward consistency model for updates through the use of an intermediate file handle that unambiguously identifies a particular sequence of bytes. We describe a strategy for a gradual transition from using URLs to using location-independent names which achieves the benefits of replication while retaining the familiar URL syntax. We also describe a service called SONAR that is intended to assist client programs in choosing among alternative locations for a file, based on a proximity measure. Finally, we describe a prototype implementation of the BFD system.

Contents: