GCJ: What were the big stories in grid computing this year?

Gillet: Well, two things strike me in retrospect. One is the union of the Enterprise Grid Alliance (EGA) and the Global Grid Forum (GGF), which ended what were basically parallel discussions occurring in separate forums. Granted, it's still a messy mix of public and private interests, but the fact that the attempt was made to bring those folks together is itself noteworthy.

And the other one, frankly, is just beginning. I think we are really only beginning to explore the interaction between what I would call scale-out compute grid - that is, using multiple machines for a common compute-intensive task - and service-oriented architecture, of distributed component software. At least, those were the things that I was thinking about this year.

I'm not sure there were any headlines about the latter trend, but the industry was certainly talking about it. I myself believe that's where grid has the most potential impact in the long run - when we redesign software for the grid architecture. I think it'll be a gradual evolution as software architecture is overhauled to deal with that kind of architecture.

GCJ: Greg Nawrocki, the Globus Consortium president, has said that he considers 2006 a make-or-break year for grid computing. Do you agree or disagree, and why?

Gillet: That implies that we're at the beginning of some take-off curve, and I honestly don't think we are. So no, I don't think of it as make-or-break. What's happened is that the term grid is getting confused. Just from a PR stance, grid has been losing attention and hype to virtualization as a buzzword. But the underlying concept has merit, and I believe it's on a gradual growth track, which means that there is no make-or-break year to be had.

GCJ: So is all the discussion of SOA and Grid 2.0 and virtualization just marketing hype?

Gillet: This is not about a technology winning or losing, this is about a contest in the public marketplace to name the overall shared, automated IT infrastructure. What's happening is that server virtualization, by virtue of being immediately applicable to the largest number of existing software applications, has more immediate impact. But ultimately, I think most customers will find distributed and virtual resources are necessary to realize the best outcomes. Keep in mind that the contest for what it's called is distinct from the contest to see who can make the technology most useful for the largest number of people as soon as possible.

See, I believe that server virtualization and storage virtualization are on the fast track to become broadly useful, and that distributed compute grids and information or data grids are on a slower track because they require more in the way of software rewrites and adaptation than the virtualization technologies do. But in the end, you do want both.

GCJ: This year, a number of people have written that to really take hold in the enterprise, grid must be standardized. What do you think?

Gillet: I think standards will help, but at the end of the day, this is about how fast the ISVs will rewrite to the technology. They're going to do that when the financial incentive for them is sufficiently compelling. So in general, standards will help, but true industry standards, the kind of top-down industry standards that we haven't seen yet, would make a much bigger difference. But it just doesn't seem to happen. It really seems to be the de facto or accidental standards-TCP/IP, for example-that have staying power. Right now, we are just stuck in one of those very annoying periods where the marketplace illogically and irrationally shakes itself out about standards.

So yes, they've got to be really important, but it's a messy process. And at the end of the day, this is about to make the ISVs happy fast, because unless they take advantage of whatever standards get created, the rest is pointless.

GCJ: All right, can we talk a bit about usage of the grid in "the real world?" In which vertical markets is grid deployed? And is it all compute grid, or has it progressed?

Gillet: Within the given scope, we get more inquiries about compute grid than data or information grid, but in both cases, its low-grade interest. We field 200 inquiries a quarter regarding various aspects of systems management, and another handful on grid issues. You have to remember, though, that we target large organizations of a billion dollars a year or more in revenue so its those organizations that aren't calling us a ton about it.

In fact, those requests are coming from the places you would expect-from people with compute-intensive, non-transactional, typically batch-oriented workloads in finance, insurance, sometimes pharmaceuticals, or engineering. So its not mainstream IT departments worrying about file servers or or running business processes, but rather places where there are specialized compute capabilities being used. So those are the industries we see and those are the kinds of uses that we see.

GCJ: What is your opinion on the Globus Toolkit and any usage that you see?

Gillet: So in my experience, there is minimal interest from commercial users - and again I refer to the larger enterprises and organizations that are our clients. My own point of view is that it looks like a useful technology for people who have for the capacity to implement it themselves, which is the hardcore audience we just discussed. I see adoption especially in the public and educational sectors, where they've got more skill than dollar, and can try to bootstrap things themselves.

GCJ: Do you have any thoughts on what the Globus Toolkit would need to get that enterprise adoption? What's missing or what's needed?

Gillet: I think what's needed is ISV support, because if the ISV packages that an enterprise is using don't support it, then it's a waste of time for them. If you go to the people who write their own code - and you find some in finance, in my experience - you find they're more interested in devoting their limited skill and brainpower to the actual business-oriented code that creates intellectual property for them, rather than trying to figure out how to implement the Globus Toolkit and integrate it with their tools. So in the end, they also seem to end up going with commercial software as an alternative to the Globus Toolkit, just because they don't see an IT advantage to them in mucking around with the Globus Toolkit.

GCJ: What are your predictions, hopes, or expectations for grid computing in the next year?

Gillet: Well, my fantasy would be to hear straighter talking about grid, but I don't expect to see any reduction in the confusion surrounding the use of the term "grid." I expect to continue to have to use my delay-of-game penalty flag every time someone uses the term "grid" and we have to spend five minutes breaking down and understanding what they mean. But my hope is that we'd all get a little better at that.

Secondly, I would like to see clearer applications of the technology, where it's useful in mundane, but important ways, like improving analytics throughput for simple enterprise tasks like report generation. So broader applicability to the enterprise especially in report generation, data preparation, and the extract transform load aspects of data warehousing, where there's a large potential to parallelize or distribute the processing associated with data transformation and preparation for analysis. So those are areas where I see immediate applicability, and those are the recommendations I give to enterprises when they call. I tell them, it comes down to identifying the elements of existing non-transactional IT processes that would benefit from distributed computing-which is what I think of as compute grid technologies.

Those are the two things I'm looking for. In general, I think it needs to be easier for the ISVs to take advantage of grid, because, to me, that's where the market leverage is - not in enabling more people to take advantage of it in the custom code they write, which is a relatively small proportion of IT spending and value.

close window