Check out the new USENIX Web site.


Web++ is a system that aims at improving the response time and the reliability of Web service. The improvement is achieved by geographic replication of Web resources. Clients reduce their response time by satisfying requests from ``nearby'' servers and improve their reliability by failing over to one of the remaining servers if the ``closest'' server is not available. Unlike the replication currently deployed by most Web sites, the Web++ replication is completely transparent to the browser user.

As the major achievement of our work, we demonstrate that it is possible to provide user-transparent Web resource replication without any modification on the client-side and using the existing Web infrastructure. Consequently, such a system for Web resource replication can be deployed relatively quickly and on a large scale. We implemented a Web++ applet that runs within Microsoft Explorer 4.x and Netscape Navigator 4.x browsers. The Web++ servlet runs within Sun Java Web Server. We currently explore an implementation of Web++ client based on Microsoft ActiveX technology.

We demonstrated the efficiency of the entire system on a live Internet experiment on six geographically distributed clients and servers. The experimental results indicate that Web++ improves the response time on average by 47.8% during peak hours and 25.5% during night hours, when compared to the performance of a single server system. At the same time, Web++ generates only a small extra message overhead that did not exceed 2% in our experiments.

Web++ provides a framework for implementing various algorithms that decide how many replicas should be created, on which servers the replicas should be placed and how the replicas should be kept consistent. We believe that all of these issues are extremely important and are a subject of our future work.