It is difficult to justify Grid computing for commercial applications - An interview with Justin Rattner - part 2

Justin Rattner: I think there is no question about that. You see some of the biggest companies including aerospace manufacturers and automobile manufacturers and how they are moving steadily into Grid computing. And that is a good example of what I was just talking about. Say, you are working on an aircraft. You might have the strucural analysis on the airframe running on a local computing resource; you might have propulsion running on another, you might have aerodynamics on yet another; but you want to link all of those thogether because you are doing total system analysis. Then you need to understand how the components interact. Those are perfect Gridlike applications. On the technical side, people are going to be moving in that direction.

It is the commercial side that is much more reluctant. There is a lot of hype about it. The IT research group within Intel has looked at Grid techniques for commercial applications, but it still has not been able to fully justify it. I think there will be some kind of combination of Grid computing and web services. If we can figure out how to intersect web services with Grid computing we may have something. But there are just a huge number of software issues that have to get worked out. This idea of the information utility as an equivalent to the power outlet on the wall has a long way to go. It requires a standardisation at a level below the application. Some people have called it service Grid. That is a kind of interesting combination of web services and computing Grids.

One of the most important features for standardisations is security. How do I distribute my corperate computing task into the Grid and be assured that my financial information, my personnel information, etc. is not leaking out? That is a very high hurdle in front of people who are advocating Grid computing in a commercial setting.

It is not hard to imagine what you do with it, but implementing it, and billing it - how do you keep track of all these cycles, how do you handle disparate storage - you are getting into so many issues. So it's going to be quite some time before standards are worked out and architectures are developed that really are useful and widely used for commercial applications.

Web services are already deployed more widely. For instance on our internal networks we support distributed training applications. We even use peer-to-peer technology. Once someone in our facility in Ireland pulls a training package from a server in the US, it stays in Ireland: the next user just gets a local copy so he does not have to use the more expensive and slower performance part of the network. We have lots of applications like this but pushing them into a service Grid where you are no longer within the Intel firewall is just too uncertain and probably to difficult to do right now.

People are active in that area, I just think it's going to be many years before this notion of information utility really comes into being.

Primeur: Is there any hardware that could help? For instance with the security part?

Justin Rattner: We are moving to a point where everything that moves over the Internet will be encrypted. That is only a matter of time. In our communication products and our processor products, we are providing direct support for those encryption standards or we are enhancing the processor to handle the encryption. But I think that is not so much the issue. It is at a much higher level: at the level of web services where the security policy needs to be worked out. The security mechanism will be there. And it is not just encryption. There are other facilities that are going to be required than just cryptographic hardware. How am I going to make sure you have the license to run a particular piece of software on this piece of hardware? How am I going to make sure that you are you? Can you give me a secure way of identifying this particular computer and that this partucular user is authorised to use it? All of this is coming, but is not a hardware problem. If it was only the hardware, we would be moving much faster. The software behind it always seems to take years.

When businesses come in and buy services that are basically provided off the Grid, you want to have some use-based fee. That is also where we have a long way to go. Again, I do not see that so much as a hardware problem but rather a software problem. There are problems with for instance IPsec-based security. One reason why it is not widely deployed is the key management problem. It is not clear whether someone will come up with a solution to key management or invent some other security structure that gets users out of doing the key management.