High Volume Web Serving

The High Volume Web Serving group has pioneered new techniques for efficiently serving dynamic content. Our work has in particularly addressed the very challenging problem of publishing large amounts of dynamically changing content while handling very large peak loads.

The techniques resulting from our research have been deployed at some of the most highly accessed sports Web sites in the world, including those for the 2000 and 1998 Olympic Games, Master's Golf tournament, Wimbledon, and the US Open Tennis Tournament. Over the last three years, these sites have maintained 100% availability despite increases of orders of magnitude in publishing requirements (up to 100Ks of changes per day) and in peak hit rates (up to 1.2M hits per minute).

Figures 1 and 2 depict the system used for serving data at the 2000 Olympic Games Web site. Figure 3 shows the growth in Web traffic we have experienced in recent years.

The group has also been active in developing Web server accelerators for fast performance. We made significant contributions to IBM's 22xx scalable Web server accelerators which significantly improve performance by reverse proxy caching. A key feature of this accelerator is that it allows dynamic as well as static data to be cached. An API is provided which allows an application to explicitly control the cache contents. Over 200 22xx accelerators were deployed at the 2000 Olympic Games Web site.

We have also developed new techniques for load balancing requests to multiple Web servers. Much of this work has been incorporated into IBM's Network Dispatcher product.

Other recent work in the department focusses on proxy acceleration, fast algorithms for storing Web data on disk, and subscription-based caching.