NCSA NEWS

In 2006, Dan Atkins was named head of the National Science Foundation's newly formed Office of Cyberinfrastructure (OCI), after chairing a blue ribbon panel on cyberinfrastructure for the foundation. Atkins discussed cyberinfrastructure with Access' J. William Bell in early 2007. Among other topics, they touched on a new vision document on cyberinfrastructure coming out of NSF, the social dynamics involved in infrastructure development, and relationships within the NSF's TeraGrid program, of which NCSA is a part.

What forces within the discipline sciences are influencing cyberinfrastructure development today?

More and more fields have intrinsically complex, multiscale problems to confront. These problems require greater interdisciplinarity, often interinstitutional organizations with distributed players... One of the reasons that cyberinfrastructure is so critical to NSF in particular is so that it can take full advantage of the interdisciplinarity across all of the fields that it supports. The supercomputer makes more and more comprehensive, multiscale simulations and models feasible and requires more collaborative activity and complementary expertise than has been required in the past. There's a positive feedback loop between interdisciplinarity and increased power of computing environments.

How do we translate that feedback loop into real dividends for cyberinfrastructure users?

Cyberinfrastructure is shifting the boundary between when it's in your self-interest to collaborate and when it's in your self-interest to compete. You're seeing, for example, more communities coming together to create community codes and share various functional models, and similar things happening around databases and production codes.

Speaking as a former dean and academic administrator, it's important that institutions make sure that they recognize these growing, intrinsic demands of science and engineering research and try to relax as many constraints in these areas as possible.

You headed up a blue ribbon panel that helped give rise to NSF's Office of Cyberinfrastructure. Now you've been in your OCI position for six months, what's struck you about cyberinfrastructure development?

[It's also shown me] we're going to have to give more attention to how to make this a true multistakeholder system and to figuring out how to be a lot more clever in how we make, leverage, and find common ground between the activities of various fields and the investments being made by various stakeholders.

It strikes me that the infrastructure in the name of the office is pretty fraught with meaning. Help us understand why that infrastructure mindset is particularly important on these sorts of issues.

You're right, we've taken the infrastructure word quite seriously as opposed to random acts of applying information technology... [This mindset] affects the role of OCI in trying to invest its money in things that are useful to a variety of fields. A layered approach where we provide resources and middleware and tools that allow communities to tailor specific resources that they need for their particular work and rapidly build customized science gateways or collaboratories or cyberenvironments or virtual research environments. We're trying to build complementarity between what we invest in and the particular, discipline-area investments that the directorates are making. And to do that systematically.

What goals does the vision document that you mentioned raise for your office and for the people building the technology?

It starts with a very bold statement of NSF's intent to provide leadership in this area. Then it talks about four intersecting components of activity and impact.

The first area, [high-performance computing,] places a high priority on science-driven, research- and education-driven cyberinfrastructure. The need, for example, for higher and higher levels of computational performance comes from the increasing complexity and multiscale nature of the problems that contemporary scientists are taking on and, in some cases, the massive amounts of data associated with doing work in that area... The commitments to supporting a multitiered system of petascale and terascale computation is pretty much well developed and a funding priority has been established for that.

With the second area, data, the points to get across are the incredible increase in the volume of data that modern science needs to contend with and the increased need for improved stewardship, coming in part from the enhanced value of data. As you know, you can extract new knowledge from data than what was originally intended, so you run a greater risk from throwing data away. You might lose opportunity.

Then the third area builds upon the first two -- virtual organizations. We're moving in the direction of more and more functionally complete collaboratories and virtual organizations, so that through a Web-based portal you can reach the primary colleagues, data, computational models, and the other things they need to do their work. When you get to that point, you quite dramatically relax the constraints of distance and time, and this ability for broad participation in authentic scientific research communities is part of this excitement in the democratization of science.

It fits with the Web 2.0 movement -- the idea that the web and the internet are not just about information. They're about participation...

The fourth area is what's being called learning and workforce development. [This is] both the application of cyberinfrastructure to enhance learning, and in particular to provide hands-on experience with science to greater numbers of people, as well as the educational and workforce demands to empower more people to use and create cyberinfrastructure.

The document talks about both provisioning the hardware and having a robust software and applications environment. Historically, that's been a challenging relationship. How do we make sure those keep up with one another?

Realistically, at least part of the approach needs to be providing mechanisms for the stout-hearted who want to do these codes initially. There's going to be a lot of hard work and hand crafting to get these codes to scale up to the peta regime.

It's also important to have mechanisms that look for common needs across fields and across disciplines and to try to extract those out and make them available so you're not constantly reinventing the wheel. So you're leveraging off the experiences and the developments of others...

The activities of the supercomputer centers and the various TeraGrid resource providers are going to be critical to that. That's really where the rubber hits the road between particular application communities and project teams that have need for or sign up to be pioneers in the migration to petascale. They're really going to be the ones who figure out how we do this.

You mentioned the concept of Web 2.0. Are there hints of attitudes within our community moving in that direction?

The open courseware movement that started at MIT [has] now extended to 100 major universities. There's kind of a feeling that resources such as curriculum and so forth should be available to the world. That's quite a significant social shift.

These fundamental shifts in social architecture and microcontribution shift the boundaries between competition and cooperation. That could actually lead to quite different approaches to how work is done...

[Web 2.0 and] those shifting paradigms of creativity and shifting notions of intellectual property...are an exemplar in the popular culture of the social impact that these technologies are having. Analogous things will happen around education and research.

One of the interesting tensions here is that infrastructure implies stability, while the Web 2.0 attitudes imply something very active.

There are sustainability and predictability issues involved in infrastructure, but you actually do want it to evolve -- in a predictable way... I think it's important to realize that we're not talking about creating the cyberinfrastructure, infrastructure for science as one homogeneous system. We're talking about a more organic process where there are all types of systems of various sorts that are being built. To make it an infrastructure we need to deal with how we get interoperability and commonality and gateways. How we accommodate the heterogeneity of systems and create a more pervasive infrastructure. We're not trying to impose a standard, we're trying to figure out a way to accommodate heterogeneity.

You can have an overall infrastructure composed of systems that are at different levels of maturity and performance. We don't want to think of TeraGrid and Open Science Grid as being competing alternatives. One is best for some classes of jobs, the other is best for some classes of jobs.

From the point of view of a functioning scientist, they should think of and be able to use both of those grids in a complementary, ideally seamless way... Let's not think of TeraGrid as one cyberinfrastructure and OSG as another. But rather a TeraGrid system and an Open Science Grid system as components of an overall cyberinfrastructure for science.

And in order to come to that, you mentioned being sure we look at it in the right way and you mentioned building the system at other levels of abstraction. What else can we at the centers-level do to encourage that, whether technical or social?

I think part of it has to do with finding a comfort zone in the space between cooperation and competition. And finding ways of all of us humbling ourselves to the collective opportunity that stands before us. The kinds of things that we're trying to work out in the context of the TeraGrid -- the balance between the autonomy of and the spires of excellence of individual resource providers but also making these sources add up to an overall system for the science community. Being positive minded about such activities.

del.icio.us Slashdot