Managing Workflow with NAMD-G

by Herbert Morgan and Kathleen Ricker, NCSA

"It started with the challenge in technology."

That's how Klaus Schulten, Swanlund Professor of Physics and Director of the Theoretical and Computational Biophysics Group at UIUC, describes the impetus for what became the collaborative scientific efforts conducted by his research group and NCSA. From that collaboration came two things: successful science and NAMD-G.

According to Schulten, one of the main challenges that the world faces today is the need to change our economy by using other fuel sources, in this case, hydrogen gas. Algae absorb light and use it to produce hydrogen gas. But there is a problem; it bubbles hydrogen gas out and permits oxygen to enter the protein thereby switching off hydrogen production.

Researchers from the National Renewable Energy Laboratory in Colorado believed that if the protein could not be changed, it could be redesigned so that it would still permit hydrogen to bubble out, but not let oxygen in. They brought the problem of redesign to Schulten and his group: Jim Phillips, senior research programmer, and physics graduate students Jordi Cohen and Anton Arkhipov. They set to work trying to discover how gas atoms travel through proteins, and how hydrogen and oxygen atoms can be differentiated so that hydrogen passes through and oxygen does not.

It was Cohen who surmised that the gas atoms do not alter the protein, but instead take advantage of an existing suitable cavity behind the protein. From that premise, one could look for protein cavities large enough for hydrogen and even larger cavities for oxygen. But this concept only posed a larger question: How does any protein conduct gases?

Whereas many NCSA users' simulations involved one large calculation, Schulten's group knew that theirs would involve utilizing multiple computing resources to perform thousands of calculations simultaneously. "And that was the moment when we realized that our savior would be the grid," says Schulten. "We did not want to use just one computer to look at one protein at a time, maybe one tomorrow or one next year. We wanted to do many at the same time, or as many as we felt compelled to examine."

However, managing workflow for such an enormous research problem was itself an enormous task. Assistance came in the form of collaboration with members of NCSA's Cyberenvironments division, including Rick Kufrin, who has been supporting Schulten's group's use of NAMD, a parallel molecular dynamics code designed for high-performance biomolecular modeling, for over a decade, and NCSA's Grid Application Support Group, led by Doru Marcusiu, Assistant Director of the Cyberenvironments division. "One way in which we distinguish ourselves is in trying to map the use of our resources to the newer tools that are part of available Grid technologies," says Marcusiu.

Following exploratory discussions in early 2005, Michelle Gower, a research programmer in Marcusiu's group, was enlisted for her expertise in Grid technologies and began working with Kufrin and the biophysics researchers to make an initial assessment of how Schulten's group was using NAMD to date. The team realized that their workflow could be streamlined significantly. "They were spending quite a bit of time doing relatively trivial management tasks, such as moving data back and forth, or resubmitting failed jobs," says Kufrin. After gaining familiarity with the existing "human-managed" tasks that are required to carry out lengthy, computationally-intensive simulations of this nature, the team identified Condor-G, an existing, proven Grid-enabled implementation of Condor, as a possible solution. "Our goal has always been that we could like to do whatever we can to improve the resources and increase productivity for the scientists, whether it involves an existing tool, a new technology, or a technology of the future," explains Kufrin.

Specifically, it was Gower who, according to Schulten, "made this idea of using the grid a workable proposition." Gower, who was experienced with Condor-G, developed a prototype system that integrated it, Globus, and other technologies with NAMD, while Schulten's group provided continuous feedback every step of the way. NAMD-G automatically handles authentication, file transfer, jobs submission, and job-chaining work. It also alerts users when something goes wrong, notifies them when the job is completed, and transfers all files onto users' computers for analysis. Furthermore, NAMD-G interacts with the queuing system and can distribute jobs to multiple sites around the world.

"One thing the prototype system does for them is to allow them to have their input files on their local machine and submit the job to the remote machine without having to understand the underlying Grid architecture," says Gower. Just two months later, Cohen was able to begin submitting jobs via the Grid. The collaboration was so successful that it became part of an NCSA demonstration prominently featured at SC05 in Seattle. Whereas technology demonstrations are often mere exhibitions that use a "mock" experiment to show what could potentially be done, this one demonstrated how NAMD-G actually helped achieve genuine scientific results.

"What needs to be stressed," says Schulten, "is that we were very demanding of Michelle because we didn't just want to engage in a grid experiment, we wanted to get some science done. And that of course made it much harder for her. The result is that we not only got our science done, but we also have NAMD-G, a program that we can offer to any other users who want to use a grid." Indeed, the NAMD program underlying NAMD-G has already thousands of registered users worldwide and this huge user group can now tap into NAMD-G.

Members of both groups emphatically agree that it was the way the collaborative process was conducted that made the project so successful. Building on a longstanding relationship with Kufrin at NCSA, Schulten's group was able to familiarize Gower quickly with the problem to be solved. Gower then identified a way in which a significant impact could be made in a short amount of time, getting it in the hands of Schulten's group as quickly as possible and working closely with them to make it robust and reliable. It's a lesson, says NCSA's Marcusiu, that he has learned from prior collaborations with science and engineering communities, such as NEESgrid. "We continue to say that building relationships between scientists and technologists is extremely important, and it often takes time. You build trust in the people you're working with in terms of competency and reliability. You also learn each other's technology or science -- or even just their vocabulary -- and IT people start to better understand what problems they're really trying to solve."

For Marcusiu, the relationship that has developed with Schulten's group is a valuable investment, and he looks forward to further possible collaboration that might bring NAMD-G, or some broader Grid implementation, to a larger community of biophysicists.

Funding for this research has been made possible by the National Institute of Health and the National Science Foundation.

FastCounter by LinkExchange