Networks could give science and scientists a new lease of life.
Are you getting all you can from your computer? Harry Hill, a British stand-up comic, once joked that to get more use from his PC, he turned it on at night and used the screen as a reading lamp.
There are many better solutions of course. He could, for instance, allow scientists access via the Internet to his idle gigabytes for experimentation, with the prospect of helping them develop a cure for cancer or a clearer model of climate change.This is what “distributed computing” is all about. Already, as researchers run out of capacity, companies such as IBM, Google and Amazon are investing in building data centres to generate “research clouds” on the Internet that students can tap into, to programme and research remotely. Now the general public can also play a role in assisting scientific research, and this is leading to a boom in distributed computing.Researchers without access to supercomputers can now pursue projects unthinkable before. According to the Berkeley Open Infrastructure for Network Computing (BOINC), the cost in hardware to install a powerful university-standard computer is about US$5 million, with annual costs of $1 million for energy use and maintenance. Most researchers can only sigh over such figures. Yet an equal capacity is obtained if research is distributed among 10,000 PCs running just 50% of the time. The cost: around $50,000.Trawling the Internet guarantees an even bigger catch. The world’s largest supercomputer, the IBM Blue Gene/L runs at a sustained processing rate of just under 500 teraflops (a “teraflop” equals 1015
“flops,” which are units of processing capacity). Participants in Folding@Home, a project that studies anomalies in the way protein molecules “fold” to produce diseases such as Alzheimer’s disease, have just passed the one pentaflop mark (1015
flops) in processing power, crunching a quadrillion mathematical operations per second.How far can open network computing go, especially if more people are eager to share their PCs? If computer processing power continues to double every two years, as has been the case, the potential is very great indeed.The way it works is simple. An Internet user registers with one or more of fifty odd projects in operation, downloads the software and lets the PC get on with the job. No need to sit hunched over the keyboard while this is going on, since the programme runs when the PC is idle. Nor does it interfere with other applications when the computer is in use. Sufficient processing power, memory and bandwidth are necessary, but most of today’s PCs meet these requirements.Users owning more powerful computers or donating more time to a project earn more “points” in a merit system created to recognise individual contributions. No money changes hands. Message boards, discussion forums and user profiles, deliberately created to foster the growth of online communities, is where participants converse and seek advice on technical problems (such as the occasionally overheating PC). At last count, the BOINC platform carried over 50 projects with a total 1.2 million users running 2.7 million computers in over 240 countries.There are projects in medicine, molecular biology, astrophysics, mathematics and climatology. One platform, Climateprediction.net, based at Oxford University, analyses the approximations of its climate change models by having thousands of Internet users run the models to see how they respond to various tweaks. This allows researchers to test the sensitivity of a model to slight changes, for example, in carbon dioxide levels or the sulphur cycle. A model runs like a screensaver, so the user can watch weather patterns unfold in one of many hypothetical scenarios.Distributed computing gives a chance to projects with broad appeal but slim financing. SETI (Search for Extraterrestrial Intelligence) has some 3 million contributors monitoring the skies for radio signals from distant civilisations.Closer to home is Africa@home’s MalariaControl.net, whose volunteers run simulations to determine the transmission dynamics of the disease. From these results, new strategies can be developed to provide mosquito nets, treatment and vaccines to afflicted populations. Africa@home is a partnership of African and international institutions, among them the Swiss Tropical Institute. Researchers at the Institute started the project with 40 computers and as many years ahead of them to complete it. But many hands make light work. In a test phase, 500 volunteer computers working over several months ran simulations equivalent to 150 years of processing time on a single computer. Thanks to the Internet, years of epidemiological modelling were completed within a few months. Similar projects are under way for lesser known, neglected diseases too.The next step is to engage not just idle PCs, but active minds. “Distributed thinking” is the flesh and blood analogy of distributed computing. Participants in a new BOINC platform called BOSSA (Berkeley Open System for Skill Aggregation) are assigned jobs demanding judgment rather than number crunching. BOINC also provides a toolkit called BOLT (Berkeley Open Learning Technology) to train volunteers.The educational possibilities are tremendous. Open networks can stimulate an interest in science at a time when the number of students in OECD countries opting for a career in science is falling precipitously. Open networks involve students in experiments of real scientific import, whose outcomes may change policy and transform our everyday lives.Openness is a defining characteristic of the Internet, and many experts are convinced that in the future Internet will depend more than ever on open, interoperable services. Already initiatives such as BOINC are moving beyond PCs, extending the grid to include other technologies. Game consoles like Sony’s PlayStation 3 use powerful cells with processing rates ten times faster than the conventional CPU (the computer’s central processing unit). Recently, Folding@home adapted its software to the PlayStation 3 console. It has been so successful that the console is expected to become the project’s main source of processing within a year.Not all research is suitable for open networks. Where personal data is involved, such as in the social and biological sciences, confidentially is a key factor. Open networks are also more vulnerable to hackers looking for any loopholes in the system.
Verification of data is another issue. Volunteers have no professional reputation to uphold, nor are they legally accountable for their data. There are even cases of users vandalising the computation. From the user’s point of view, how can one be sure that the software downloaded for a particular project will not damage a PC?Intellectual property rights are another concern, as scientists (and laboratories) worry that their work may be stolen unless it is under patent. It is a divisive issue and policymakers are addressing this concern.The message for policymakers is clear though: a major barrier for open source research is the lack of high-speed connectivity in many countries and the high cost of local telecommunications hook-ups due to national pricing structures. It also demands a fresh mindset. Policy will have to adapt to this trend by creating incentives that encourage openness and co-operation in the exchange of knowledge without sacrificing scientific integrity, value or personal security.Far from being a threat, the popularity of distributed computing offers new opportunities to professional scientists. The Internet may speed up research and offer new resources, though the supply and processing of information from the network demands great scientific judgement, skill and knowledge. The demand for science and scientists could well grow. In short, thanks to the Internet, scientists cannot just catch public interest, but harness their computer power too. LTReferences
OECD (2007), “Networked Research and Research-Intensive Business Scoping Outline”, DSTI/ICCP/IE(2007)16, 14 November 2007.Find out more about BOINC initiatives at http://boinc.berkeley.edu/index.php
See also www.worldcommunitygrid.org/ projects_showcase/viewResearch.do
For more, contact GrahamVickery@oecd.org
©OECD Observer No 268 June 2008