Saving Lives with P2P
With distributed computing, scientists improve vaccines, research cancer, share medical recordsby Bruce Stewart
Dr. Derek Smith is using advanced computer modeling to help develop more effective influenza vaccines. Leading a project for the World Health Organization, Smith isn't running his projections on a multimillion-dollar supercomputer, but on more than 10,000 desktop computers spread across the Internet, using Popular Power's distributed-computing software.
The flu vaccine project is the latest example of a scientific problem that is being tackled by a distributed-computing model. Peer-to-peer technology first gained recognition for its ability to harness unused processor cycles to work on large problems with the SETI@home project, which enlisted thousands of Internet users to install a special screensaver that evaluated data collected in the search of extraterrestrials.
This type of computing model involves installing client software on many networked computers, and doling out small chunks of a larger problem to the clients when the computer is not busy with other processing. The new breed of distributed computing systems take advantage of the unused resources of networked computers, doing their work in either a screensaver mode, or in the background while other tasks are running.
A key difference in the SETI@home project and the distributed platforms discussed here is that these commercial ventures aren't just sending data to a program that was previously installed on a client, like the SETI screensaver, but can now send code as well as data, so that the clients can work on a variety of different problems.
Companies such as Popular Power, Parabon Computation and CareScience are using P2P technology to help solve serious medical problems. Distributed Science is using a P2P application to further the research of a Swedish scientist trying to find safer ways to store radioactive waste. The medical projects showing up in P2P applications will showcase the power of the technology and give some needed help to difficult scientific problems.
Popular Power: Influenza Vaccine Project
Popular Power released its P2P client, called Worker, in April. This was the first commercially released distributed-computing client designed to utilize the Internet, by sending code and data to nodes that make their computational resources available for large projects.
The company launched its network with a nonprofit medical research application that studies influenza vaccination. The project attempts to help researchers find better influenza vaccines by simulating immune system responses to different vaccines and influenza strains. It is being led by Smith of the World Health Organization's National Influenza Laboratory in Erasmus University, Rotterdam.
Smith has constructed a computer model of the human immune system, which he "injects" with different vaccines and tests their efficacy against different viruses, according to Debbie Pfeifer of Popular Power. "His work is helping to create better strategies for selecting the influenza vaccine, which changes every year because the influenza virus mutates."
He then bundles several individual simulations and sends them as a "task" to member computers. He provides parameters for each simulation sent -- instructions for what will happen in the simulation running on each node, using different variables such as vaccination history, and virus strains. When an individual computer is finished with its simulations, the results are sent back to Popular Power, and new simulations are sent.
The results from all the nodes are collected and collated by Popular Power and the results are then passed back to Smith for analysis. "Our system enables Dr. Smith to run many more simulations with more variables at a faster speed than he was previously able to do," Pfeifer says.
The Worker client works either as a screensaver or in background mode, where it can take even more advantage of a computer's unused resources. The client is a combination of some native code and a Java runner -- the Java core of the system. The native code's responsibility is to install the program onto the computer and act as a screensaver and mediate interaction with the OS. The Java client handles all of the networking and job execution functions.
Nelson Minar, co-founder and CTO of Popular Power, points to size, reliability and scalability as the key challenges in building a reliable computational platform that makes use of clients all over the Internet.
"The biggest challenge is building a large reliable, scalable system on the Internet. The Internet itself is flaky: links go down, data gets lost, etc. And the clients are flaky: the user turns off the screensaver, or the OS crashes, or the floating point unit has a bug, or the whole system is turned off by a California rolling blackout. Our challenge is to take all these unreliable raw materials and turn them into a reliable resource."
Minar says the sheer number of clients give them redundancy, which allows the system to detect and correct failures. The system's "backoff strategy" is a good example -- if a client can't reach the server it keeps retrying, but does so with an "exponential backoff timer" so the server never gets overwhelmed with a crush of client connections.
With more than 10,000 clients running the software, Smith's influenza project has run millions of experiments since April. Popular Power hopes this serves as a good reference application to demonstrate the strength of their platform. Popular Power also has a beta client installed at a major pharmaceutical company that is evaluating the software for drug design applications.
Parabon's Compute Against Cancer
Much like Popular Power, and the SETI@home project before it, Parabon Computation offers a distributed computing platform that can utilize resources across the Internet.
While Popular Power brought the first client to market, Parabon was the first to release a complete system for this type of distributed computing, including client, server and SDK and API components.
Cancer research is the beneficiary of Parabon's technology -- the company has sponsored the Compute Against Cancer program. Projects in this umbrella program are working to understand and reduce the side effects of chemotherapy, study the structure and behavior of cancer cells and create better ways to screen new cancer drugs.
The Parabon application uses exhaustive regression, a statistical technique that makes it possible to look at each combination to identify which factors improve quality of life. This technique attempts to turn a massive amount of data into helpful therapeutic information for cancer patients.
Parabon's client software is called Pioneer and is Java-based, as is the server component. The system has been in development for about a year and a half, and there is now an Internet-based platform called Frontier, and an enterprise version called Frontier Enterprise.
Parabon's High-Performance Computing Lab has developed a new application that uses Frontier to conduct protein sequence comparison. Applied to publicly available protein databases, this approach will result in the most comprehensive and thorough comparison ever performed, according to Parabon. This research could give scientists and medical researchers a better understanding of the mechanisms at work within normal and malignant cells, clearing the way for developing more effective cancer therapies.
Jim Gannon, CTO of Parabon Computation, says they are talking to potential customers with 50,000 to 90,000 workstations in the enterprise. The computing power that sits idle in these large organizations is staggering.
Gannon points out that the biggest supercomputer on the planet, IBM's ASCII White, has been equated to 30,000 to 50,000 desktop computers, so large companies have even more power just sitting there.
When asked whether any concrete progress had been made in cancer research that could be attributable to Parabon's program, vice president Mark Weitner couldn't really brag. Parabon's work with the National Cancer Institute is being submitted for publication, so for now mum's the word.
Pages: 1, 2