In-Depth

Q&A With Sun’s Grid Guru

Dr. Wolfgang Gentzsch, director of grid computing, discusses the distributed computing technology

• By Stephen Swoyer
• 02/04/2003

Prior to that, Dr. Gentzsch was a professor of applied mathematics and computer science at the Technical University of Regensburg, Germany. In the past, he has consulted on supercomputing projects for IBM Corp., Cray Research, and the erstwhile Digital Equipment Corp., among others.

In July 2000, Sun acquired Gridware, and, with it, Dr. Gentzsch. Since that time, grid computing has emerged as an important strategy for Sun and for other industry players, particularly IBM Corp.

Last week, the White Rose Grid, an enterprise area computational grid that connects the Universities of Leeds, Sheffield and York in the United Kingdom, officially went online. White Rose is enabled by means of Sun’s Grid Engine Enterprise Edition software, Sun’s Grid Portal Development Kit and by version 2.0 of the Globus Toolkit.

We recently sat down with Dr. Gentzsch to talk about Sun’s involvement with grid computing.

ES: Tell us about the White Rose Grid.

WG: The White Rose Grid combines compute resources in Leeds, York and Sheffield. These are basically four clusters, very large cluster grids of Solaris-SPARC and Linux-Intel, and the key difference technology-wise is that now the Globus Toolkit is involved on the upper most level. Basically, this [Globus Toolkit] provides a secure means for communications. Underneath, you have our Grid Engine Enterprise Edition, and that is responsible for managing the grid computing resources, and it also enables policy management.

ES: You’ve described a computational grid that connects geographically dispersed university campuses. This isn’t necessarily new. What’s of interest here for business-class customers?

WG: This White Rose grid announcement is the start of a whole series of announcements which will follow about real global grids, which Sun is helping customers to build with Sun grid technology. So as to not frighten or irritate our customers, we started with small grids, grids that can easily be built with core technology like the Sun Grid Engine. So this evolutionary strategy started at the low level and went to the enterprise level.

We are now announcing more and more global grids. White Rose is one example of that. The difference between the former grids and the global grids are that it’s a virtual organization that spans distributed partners, be it distributed university or research vendors, or distributed commercial or research partners.

ES: And this is done securely over the public Internet?

WG: Yes. [These universities] are 50 or 60 miles away from one another. [The grid] uses MyProxy, [which is] a specific technology that provides a server with client-side utilities to store and retrieve dedicated x.509 credentials, so this is a credential pool and it’s delivered via the grid security infrastructure (GSI) so that you can store and retrieve the security credentials specifically to decide who is allowed to access what.

There’s no reason why the same architecture cannot be applied country-wide.

ES: Last week, IBM announced new grid computing bundles designed for specific vertical industries, e.g., aerospace, automotive, financial markets and government. It also touted at least one customer, Charles Schwab, that’s using IBM grids to support a business application. How does Sun propose to compete against IBM? And can you point to any examples in which Sun computational grids have been deployed to support commercial applications?

WG: The difference between Sun and many other vendors is that we own all of this technology, and that we basically started developing it, and gave it back to the community, and we have a very large involvement in the grid open source community. Besides the over 7,000 commercial production grids which we have helped to build in the last two years, there are thousands of downloads of the software on the open source side.

In November, we announced the Sun Grid Portal, and we immediately put it into Open Source. We have grids for 10 different vertical markets, and we have on the order of two dozen pilot projects ongoing with this Sun Grid Portal technology. It’s nicely integrated with our portal engine, and also has a grid to the Sun ONE net services to use it more with commercial grid environments.

We have successfully deployed grids for commercial applications. Ford [Motor Co.] has a Sun grid that does modeling to help its Powertrain group design better, safer cars. With Ford, we built a grid with 1,000 processors, very flexible, with two main tasks, one for basically designing things, one for computing things, simulating things, and this same architecture we have replicated in many other MCAD environments. We have grids with Cognigen [a data analysis and consulting firm that serves the pharmaceutical and biotech industries], analyzing the effects of drugs on a population.

ES: IBM is probably the most platform-open vendor in the market today. Sun obviously supports SPARC-Solaris, and you’ve embraced Intel-Linux, but how would you compete in a commercial account that wanted to deploy grids based on, say, AIX?

WG: We realize that people will want to use what they already have. They will make their own decisions about what to use. People don’t simply say: “I want to buy a Sun Blade 2000.” They will tell you: “I have this problem to solve, this is my current architecture, I don’t want to remove it. How can you help me?”

There is absolutely sufficient reason to use pure Sun technology, but if people already have something different, then we work with them. Tokyo University, for example, has a grid with 777 [Sun] systems. They are using Solaris and SGI IRIX systems in a mixed environment, so the 777 [computers] I mentioned are only Sun systems, and there are another 500 SGI machines. This is all heterogeneous, We don’t simply or naively throw out [the] competition’s hardware. We help our customers.

ES: What’s going to drive the adoption of grids in enterprise environments?

WG: Grids are to be especially sexy in this very bad economy. You know these numbers that resources are utilized only on average between 15 and 30 percent? On the other hand, when you install a grid, like we do with our SPARC simulation farm, this enterprise grid is utilized 98 percent on average, so you see that this is a factor of three to four [times more utilization]. This enables people to get much more use out of what they already have.

Another reason is the seamless sharing of resources. In the early days of the Internet and World Wide Web, you didn’t share resources, you exchanged information, and that was all. But now you share resources for larger projects, or making computing more efficient. The [Sun] Grid Engine transparently allocates all of these resources.

Finally, with something like the Sun Grid Engine Enterprise Edition, you can explicitly have the information about who used what, when; then you can allocate expenses to them. You can make [business units] accountable for the resources that they use.