“Commercial grid is no longer just a vision”

Michael Fehse
Michael Fehse

Why are you so sure that grid computing will become generally available?

Fehse: In the technical-scientific environment, grid systems are quite widespread. They consist of unused computing resources from different establishments that can be combined as a cluster. Seti@Home, the initiative for investigating extraterrestrial intelligence, is currently the largest cluster of computers. In Hungary, for example, a supercomputer has been put together that consists of a thousand PCs which are used for education purposes in a university environment during the day, and form a supercomputer at night, when they are not busy. In the technical-scientific sphere, there are already more than 10,000 grids. They function well, but do not meet business and economic requirements. We want to change that. With the “next generation grid” or “commercial grid,” we are pursuing the idea of transporting scientific grids into a commercial environment.

What is still missing on this route towards a commercial grid?

Fehse: Conventional grids are trimmed for “number crunching.” This means that they can process huge volumes of data in record time – for example to simulate processes on or in the sun, to calculate protein molecules, or evaluate trials in the particle accelerator of the European research center CERN, which is now dealing with data volumes of two- to three-figure petabytes. A petabyte is one thousand terabytes – a one followed by 15 zeros. One disadvantage for a commercial service, however, is that usually, very exotic network protocols such as Inifiniband or Myrinet are used. Nor are these applications based on clear service level agreements; “cycle scavenging” is used instead, describing a net search to obtain CPU cycles from anywhere.

What do you want to offer company users based on commercial grids?

Fehse: Disaster recovery, redundancy, downtime security – those are the most important promises that we want to keep. The commercial grid can enhance the attributes of the grid that are not so important for scientists. This relates to the flexible architecture. Grids are created in such a way that they regard the downtime of a component as normal. Thanks to their architecture, they are prepared to react extremely quickly to a downtime or loss. If capacity is lost somewhere, even on a large scale for example as a result of a serious fire, or in a typical scenario such as a digger cutting through a cable, the grid automatically organized a replacement in the network. Because everything is based on IP addresses, the replacement computer could easily be in New York, without the high degree of effort that is involved today.

What do you imagine a business model such as this would be like?

Fehse: The system will be based on rules that are derived from service level agreements. Users and customers are provided with their services. They get IT resources, usually without having specifically ordered them, whenever they or their applications need them. As a general rule, there will be three partners in this transaction. First, the users who want to use the computer resources. Then the agents or coordinators. And finally the resource providers. The party that provides its resources does not automatically have to be the party that offers the applications and service. Projects are currently under way together with the EU and large companies to define exactly where these interfaces will be.

Where would the resources for a commercial grid come from?

Fehse: Large companies have a huge number of unused PCs. The customers supported by T-Systems alone have resources amounting to 1 million PCs. But grids are not just restricted to PCs. At T-Systems, we are also able to use our 30,000 or so servers. And other systems, including very mixed ones, could also be bundled. Initial trials with heterogeneous grids are going very well. This means that we do not even have to “cluster” uniform resources in the sense of just blade centers, or just mainframes, or just PCs.

How will grids cope with transactional applications?

Fehse: Some research and development work is still needed here. In special cases such as web shops and relational databases, this is no longer a problem. But in complex applications, which involve several different types of transactions, such as in ERP solutions, for example, we still have some way to go. The main problem in these transactional systems is that some jobs run continuously. In the scientific environment, on the other hand, everything can be foreseen. In a transactional system, it is not possible to predict the use of resources. This is a problem.

Can you guarantee security and the protection of personal data when data is processed in the grid in open standards such as XML?

Fehse: Six months ago, I would have said that would be a showstopper, but we have come a long way since then. The main problem is that classical screening technologies do not see the document, just bits and byte packages. Classical signature-based or traffic route-based intrusion detection procedures or other firewall functionalities fail completely here. This means that you have to look at the whole document, which may even be partially encrypted. In this case, how can the encryption codes be handled? We still need to clarify this. We have embarked on a project with a number of firms and several large agencies that specialize in this area.

What are the business expectations?

Fehse: The commercial grid is no longer just a vision. With it, we have developed a completely new operational model, and can provide resources autonomously, automatically, and dynamically. We assume that, on a broad scale, grid technologies will improve the quality of our services and will usually appear in hidden form in them. We believe that this technology is so ground-breaking and fundamental that not to use it would be a similar mistake to not using the internet. And consequently, we do not even need to calculate how much more business we could generate with grid computing; we merely need to look at how great the danger would be of not using it.

What timeframe can we expect?

Fehse: I expect the first marketable commercial grid systems to be available by the end of 2004. I think we can expect to see a wave of grid service offerings by mid 2005. The services will then be generally available by the end of 2005.