Although the Web is widely used today, for it to be truly effective part of daily work, it must provide both dependable availability and performance. Today, the Web often suffers in both these categories because of overloaded servers, unreliable networks, and mobile (and therefore intentionally disconnected) machines.
Replication has the potential to improve both availability and performance in the Web. Replication does both of these things by locating pages ``closer'' to the end-user.
Consider availability: network interruptions are an all-too-common occurrence in today's wide-area networks. Businesses often expect 7x24 reliability out of their computing services; lack of such reliability in the Internet discourages its use in these roles. Document replication can addresses network reliability issues by placing copies of important data ``closer'' where reliability can be guaranteed.
An increasingly important special-case for availability is mobile computing. A portable computer is often intentionally disconnected from the network. Although a number of wireless approaches to communications have been proposed, but cost and geography often prevent their use. Data replication on portable computers can again address this availability problem.
Performance is the other major motivation for replication. Even when wireless networking is both affordable and available, bandwidth may be less than is required. Data replicated on the local disk is immediately accessible.
Performance over wide-area networks is also a concern. Link saturation and packet loss (either of a company's Internet gateway or an transcontinental link) are a frequent problems. Even when bandwidth is available, the speed of light requires about 30ms to cross the United States and nearly 300ms if via satellite. In these cases access to a closer replica can substantially improve performance.
Finally, a single server can prove to be a performance bottleneck. Replication can distribute load across several servers, eliminating this bottleneck.
How can we quickly find the ``nearest'' replica in a variety of environments?
Page maintainer: John Heidemann
Last modified: Mon Jan 6 10:48:20 1997
Copyright © 1996 by USC/ISI