The Last Word

In this chapter, we considered the design of a distributed work manager. This type of application promises vastly improved performance, but introduces new complications and requires much more work than a stand-alone application. However, the benefits increase as the pool of available workers increases and the task load mounts. SETI@Home, the largest public-distributed computing project in terms of computing power, reached a record 71 teraflops per second on September 26, 2001. By comparison, the fastest individual computer in the world, IBM's ASCI White, runs at 12.3 teraflops per second. ASCI White costs over $100 million, while SETI@Home cost an estimated $500,000 to develop.

Distributed computing is highly dependent on the problem domain. Some approaches work well for certain types of problems, and some tasks are inherently more suited to distributed computing than others. Most distributed supercomputers have their own individual approaches, which are customized based on the task and type of data. In the future, it's likely that broader standards and a consistent framework will emerge from communities such as the Global Grid Forum (http://www.gridforum.org) and Globus (http://www.globus.org). For a list of some current large-scale distributed applications with their performance information, visit http://www.aspenleaf.com/distributed/distrib-projects.html.