the economist magazine (dated 28 july 2000) reports on a large distributed computing project called SETI@home (most of you of course already know about it, and many are part of it!).. .. full text of article below. THE most powerful computer in the world is sitting neither in a secret military base, nor in a university laboratory, nor even in a garage in Silicon Valley. It is, in fact, nowhere in particular. Part of it may even be on your desk. The computer in question is a "distributed" device that consists of over 2m separate machines sprinkled around the Internet, all running a screen-saver called SETI@home. This piece of software downloads chunks of data from the Arecibo radio telescope in Puerto Rico and, when the machine it is installed on is not doing anything else, scrutinises them for evidence of signals from alien civilisations, sending the results back to a central clearing-house. So far, no aliens have been found. But in the 15 months since the project's launch, the machines running the SETI@home software have put in a total of 345,000 years' worth of computer time. These machines are collectively the equivalent of a computer operating at around ten million million calculations a second, about ten times faster than any conventional supercomputer. All of which has got a number of people thinking: why not harness the power of distributed computing for commercial gain? The idea would be to farm out large computing tasks to thousands of individual PCs. Vast computing power could thus be provided on demand, and the individual members of the collective paid for the use of their machines-which would probably have otherwise been sitting doing nothing. Many a mickle makes a muckle There are, inevitably, several problems to overcome before this something-for-nothing idea can actually be made to work. First, there is the question of getting the software to run on as wide a variety of computers as possible, so as to maximise the number of machines available. Second, there is the issue of security. Will people be prepared to farm out potentially sensitive work to an anonymous collective? And third, not every kind of large computational problem can be broken up into the sort of discrete chunks that can be processed by individual machines. Nevertheless, over the past few months several new firms, each with different solutions to these problems, have popped up to exploit what they believe will become a lucrative market. Steven Armentrout, chief executive of one such company-Parabon Computation, based in Fairfax, Virginia-points out that large organisations often need computing power in bursts, but that supercomputers are designed to provide sustained power. Using lots of small computers together could provide these power-bursts more cheaply, since a company would not need to leave an expensive piece of "big iron" lying around unused for much of its life. So far, Parabon has recruited over 3,000 users for its software. To deal with the problem of getting that software to work on different sorts of machines, it is written in Java, a programming language specifically designed to do just that. To respond to security concerns, all network traffic is encrypted, and each machine handles only a tiny fraction of any client's data. And Parabon aims to answer the question of which computing problems are appropriate for distributed solutions by focusing initially on financial modelling and on the search for genes in raw DNA sequences, both of which clearly are. The firm is currently negotiating with potential clients in the fields of finance, pharmaceuticals and bio-informatics. Pay scales for participating computer owners will be determined once a client has been signed up. If they wish, owners will be able to donate their earnings to charity. Popular Power, based in San Francisco, started testing its distributed-computing software in April, and now has around 5,000 users. This test version, again written in Java, is now being used on a non-profit basis to conduct research into the relative merits of different influenza-vaccination strategies. But the company's boss, Marc Hedlund, says it is close to signing up its first paying customers. He also hopes to do deals with other firms that already have large numbers of computer owners signed up-Internet service providers, for example, and online retailers. In return for running Popular Power's software, Mr Hedlund suggests, owners might get free Internet access, or discounts on online purchases, rather than actual cash. A similar strategy is being pursued by Distributed Science, based in Toronto. The firm has already accumulated what one executive calls a "mercenary army" of 40,000 users to handle its first paying job, and its software is now running an experimental simulation to evaluate the design of nuclear-waste containers. Chris Harrison, a co-founder, says Distributed Science has chosen not to use Java on performance grounds, but will release different versions of its software for different kinds of computer. So far, none of these firms has gone from the testing to the money-making stage. But there are several reasons to believe that distributed computing will be a viable idea. Dr Armentrout points out that IBM's ASCI White, the fastest computer in the world, has a power equivalent to a mere 30,000 desktop machines. There are 100m computers connected to the Internet in America alone. As fixed connections (such as digital subscriber lines and cable modem links) become more popular, many of these machines will be online around the clock, even though they are doing nothing most of the time. Distributed computing would allow their wasted processor cycles to be put to good use. Mr Hedlund notes that, as well as exploiting unused processing power, distributed computing could also harness unused network capacity. Search engines, for example, find it difficult to maintain an up-to-date directory of the World Wide Web, because it takes them a month or so to "crawl" round every web article to see if it is still there, and whether its contents have changed. It would be more efficient to distribute the job of crawling to thousands of machines around the Internet, which would then tell the central search engine which pages had changed. The next logical step in distributed computing will be to enable members of a collective to communicate with each other directly, thus forming a more efficient "virtual machine" that would be able to perform far more complex calculations than is currently possible. Large firms might also wish to make use of the technology over their internal networks to exploit the collective power of their desktop machines, perhaps to perform complex calculations overnight. That would get rid of many of the security problems associated with letting private data out into the wider world. There is, in other words, plenty of scope for innovation. The underlying principle is that many hands make light work. Proponents of distributed computing hope that many hands may make a profit, too. ----------------------------------------------------- The mailing list archives are available at http://lists.linux-india.org/cgi-bin/wilma/linux-delhi/
