7.2.5 Latency Tolerant and Intolerant

		10.2 New Opportunities

		Historically, Beowulf-class computers have been used primarily for science and engineering applications or as experimental test beds in educational settings. But their low cost, scalability, and generality provide a wide array of opportunities for new domains of application. Some of these can be accomplished now with existing hardware and software technology. Others require the development of new and innovative software infrastructure. Some examples of these new application opportunities follow.

		10.2.1 Databases

		Commercial software relies heavily on the management and manipulation of data in complex databases. At the time of writing, major vendors of widely used commercial database software systems including Oracle, Sybase, Informix and Objectivity have announced their support of the Linux operating system. Beowulf presents the opportunity to support distributed databases for coarse grain searches and multiple transaction processing tasks for multiple simultaneous users.

		10.2.2 Web Servers

		One of the most rapidly growing responsibilities of local systems is hosting World Wide Web pages and servers including search engines. In the last few years, the web alone has caused Internet bandwidth usage to increase by more than an order of magnitude in certain areas. This is due, in part, to the heavy emphasis on image data that is a significant component of many web sites. With appropriate server support software, Beowulf can provide a low cost, high performance, scalable platform for future web servers.

		10.2.3 Dynamic Interactive Simulation

		Dynamic interactive simulation (DIS) is a class of applications that present multi-user shared simulated environments with real-time response. DIS systems have been applied to everything from war gaming to multi-user video games. DIS requires complex scenario simulation, fast response time and sophisticated user interfaces. Beowulfs can be a low-cost, scalable platform to stage complete simulations or they can act as computational elements in larger, more dispersed simulations.