Last week, the UCSD division of the California Institute for Telecommunications and Information Technology (Calit2) and the J. Craig Venter Institute announced that they would collaborate to decipher the genetic code of the world's marine microbiological communities. This project, the Community Cyberinfrastructure for Advanced Marine Microbial Ecology Research and Analysis (CAMERA), will use the OptIPuter model developed at Calit2 as the architecture for its computational resources. According to Larry Smarr, Calit2 director and principal investigator for the CAMERA project, this work will represent “the first persistent application of the OptIPuter.”
Named for its use of Optical networking, Internet Protocol, computer storage, processing and visualization technologies, the OptIPuter is an infrastructure that links computational resources over optical networks using the IP communication mechanism. The OptIPuter's central architectural element is optical networks, not computers. The goal of this architecture is to enable researchers who are generating large volumes of data to interactively visualize, analyze, and correlate their data from distributed sites.
Bringing it Together
“The OptIPuter from the beginning was driven by scientific applications — in particular biomedical imaging, earth and ocean science,” explained Smarr. “But it was really more for individuals who had some large datasets that we could play with — the scientific research project itself was the OptIPuter as computer science.”
The OptIPuter project, funded by the NSF, was in its fourth year of a five-year project. Smarr knew that he would have look for an opportunity in the scientific community to take the project to the next level. Coincidentally, the Gordon and Betty Moore Foundation was looking for just the type of computational resource provided by the OptIPuter.
“It just fortuitously happened that the Moore foundation asked me to become the principal investigator for [the CAMERA project] and for Calit2 to put this together at just about the time that we have would be looking for such a project in the scientific community and go the next step — proof-of-principle,” said Smarr. “So this is really good timing, from our point of view.”
It wasn't all just luck, however. Larry Smarr, as one of the luminaries in the field, is well known for his contributions to the information technology community, from his early involvement in the original Mosaic web browser at NCSA to his current work as the founding director of Calit2. David Kingsbury, the science program officer at the Moore foundation, was well-aware of Smarr's work.
“David has been driving computation biology for 20 years – first at the NSF, then with Chiron Corporation and then with the DOE for awhile,” said Smarr. “I've known about him since my NCSA days and we'd interacted in the past. It turns out he was also selected by the UC office of the president as one of the team of reviewers for the Calit2 proposal back in 2000. So he was aware of my track record both at NSCA and then at Calit2.
“David was looking for a place that wanted to live in the future — beyond the leading edge of technology, but was driven by science. He knew that one of the four major application areas for Calit2 was digitally enabled genomic medicine. As a result we had collected a number of leaders at Calit2 in computation biology and bioinformatics. We had a in-house capability in both leading edge information technology and computational biology. So this seemed like the right group for him.”
Prior to Calit2's involvement, the Moore foundation had been funding the Venter Institute for several years to collect marine microorganism and sequence their DNA. The Institute's marine expeditions were used to collect water samples from a wide variety of locations around the world. While at sea, the microbes are filtered from the seawater, and then frozen for transport back to the Venter labs in Rockville Maryland. After the samples are brought back to the Institute, the real fun begins.
“They shotgun sequence the whole lot so you end up with this very complex genomic map of a whole community of microorganisms, that have adapted to this local environment,” said Smarr. “Each sample can contain thousands of species.”
But that's the easy part. They also need to correlate what the environment was like when the sample was collected — the temperature, pH, salinity, as well as the local ocean context (obtained from NASA satellite images).
“So from a computer science point of view, you need to have a broad set of data types, large volumes of data, and a whole set of software tools that have to be applied to that data to get the science out,” explained Smarr. “There really wasn't any existing science complex that was set up to do this. So they literally wanted us to architect a new kind of science data server that would not just satisfy this particular metagenomics project but would be a 21st century architecture that would have five to ten years of legs on it.”
Smarr envisions the expansion of this technology in two directions. First, he believe this project will help to accelerate metagenomics research, not just for the marine environment, but for other microbial ecosystems as well. Second, as they prove their ability to support distributed teams in these virtual “collaboratories,” Smarr expects to see the technology translated into many other scientific disciplines, such as astronomy and chemistry.
“What is exciting about this is that it's taking both frontier science and combining it with frontier cyberinfrastructure,” said Smarr.
Beside basic scientific discovery, there are several of potential applications for metagenomic research. According to Smarr, there are a number of companies that are already looking at marine microorganisms for new drugs, the way they have with soil-based microorganisms. There are also exciting biofuel applications that are being considered, for example the production of hydrogen and ethanol as fuel sources from microbial metabolism.
“Craig [Venter] is particularly interested in hydrogen fuel. One of the things some of these marine microbes do is to produce hydrogen as a waste product,” said Smarr. “So there's this notion of creating synthetic bacteria that have a specific engineering or health application in which you would insert the gene sequence for a certain kind of activity into a worker microbe. That's a whole new industrial revolution.”
Smarr also projects how the technology can be applied directly to other microbial ecosystems. For example, the microorganisms inside of the large intestines were recently shotgun sequenced by Stanford researchers. Soil microorganisms, the source of many drugs, such as penicillin, are another likely target for metagenomics. Even airborne dust particles can be biologically active and are currently being studied in relation to the mold problem caused by the aftermath of Hurricane Katrina.
“What we expect to do is reach out to these other scientific projects that are developing microbial metagenomics and see if they're interested in working with us on this new architecture,” said Smarr. “I think the biological community has been calling for us in the computer science community to help them with their exploding data problem. We want to become more visible in that community simply because we think this answers what they've been searching for.”
The OptIPuter Paradigm
The OptIPuter model is based on the ability of optical networks to move data around at speeds of tens of gigabits per second over dedicated lambdas. Significantly, the increases in optical network bandwidth and storage capacity are outstripping the increases in CPU performance. As a result, “Moore's Law” is not driving information technology the way it used to (ironic when you consider that Gordon Moore, the originator of “Moore's Law,” is now funding this project through his Foundation).
The OptIPuter exploits the enormous bandwidth of fiber optic networks to link distributed computer and storage resources. With the recent expansion of National LambdaRail as the optical backbone for cross-country connectivity, Smarr believes we're entering a critical stage for technological change.
“This is a one-in-twenty-year transition point,” said Smarr, “going back to 1985, when the NSF built the first backbone for the shared Internet. Now National LambdaRail has built the first backbone for the unshared Internet. At present, there are about two dozen state and regional optical networks that are interconnecting to National LambdaRail. The campuses are beginning to put fiber optics into their actual laboratories, and connecting these to the state and regional optical networks which are then connected to National LambdaRail.”
The adoption of computer clusters as a standard tool for high performance scientific computing is another factor that is driving the transition to faster interconnects.
“What's the natural I/O speed for a Linux cluster,” asks Smarr? “The average sized cluster at the the University of California is about 32 nodes. The typical network card is Gigabit Ethernet. So your Linux cluster wants to talk to the rest of the world at 32 Gigabits per second. But even using Internet2, you're lucky to get more than 50 Mbps. You're off by two or three orders of magnitude. We think we're connected to the data, our colleagues and remote instruments by the Internet. In fact, we're incredibly cut off from them. We live in little data islands and compute islands.
“So that was the fundamental insight that led us to work on these optical networks. It wasn't that optical networks were cool and we were looking for something to do with them. It was that the scientific community had decided on Linux clusters as their standard and they're natural need for a wide area network was clearly in the gigabits and tens of gigabits per second range. So we looked around for a technology that could provide this and found that the telecom industry had evolved to the point where the natural data flow on their individual lambdas was 10 gigabits per second.”
The Science Server
As part of the CAMERA project, Calit2 will partner with UCSD's SDSC to develop the science data server complex, which couples the Calit2 and SDSC middleware, compute, and storage capabilities with the TeraGrid computing facility in a Service Oriented Architecture. This will enable computing resources to be applied to a range of tools to tackle the computationally intense questions derived from the metagenomic data collection.
SDSC's Philip Papadopolous is the co-PI for the CAMERA project and is the architect for the data computation and storage server. According to Smarr, this is going to be a very advanced server. Calit2 is working closely with Dell and Sun Microsystems to build one of the most advanced compute and storage systems ever put together. At completion, it should have at least 1000 processors and contain several hundred terabytes of replicated data storage.
“What's most exciting is that at the center core of this compute and storage complex is not a computer it's a 10-gigabit optical fabric,” said Smarr. “Everything is built as peripherals around that. We're working with the vendors to get extremely high I/O storage.”
When the CAMERA science server is developed, it will appear as a network appliance, albeit a very powerful one. This could be one of the most important milestones for the TeraGrid. Here's how Smarr explains it:
“This is the first science data server that has been architected to direct-connect to your local cluster through the National LambdaRail. What we've done with this server is make it the first TeraGrid appliance. In other words, we're linking directly into the TeraGrid lambdas from our science server. So as a user, when you connect to the science server, it now appears to be just an extension of your local cluster. Over the next few years the TeraGrid will expand to tens of thousand of processors, so you'll get orders of magnitude increases in power by plugging into the TeraGrid. It should all appear as if it's in your laboratory. And that's the vision!”
To learn about another OptIPuter application, read Optical Race, the next feature article in this week's issue.