Binary Freedom: There's been talk about the clustering potential of Linux for a while- we've had Sun employees working on clustering leave to form the Linux Clustering Cabal, and various other Operating Systems have been ditched in favour for something with more computational power and less kernel locks... how scientific was the process of choosing Linux for the cluster? Is there some property of Linux that's going to help you out here?

Dan: Linux was chosen for a number of reasons. The first is the level of acceptance that it has gained in the scientific computing communities. This simply means that it is possible for users of NCSA's cluster systems to also have their own cluster that is compatible with the large-scale NCSA clusters. This gives the user the capability of migrating between the systems with minimal effort.

The second is the availability of Linux on commodity hardware from a large number of vendors. Traditional supercomputers have proprietary hardware and software, both of which are very expensive to purchase and maintain. Commodity hardware and software systems allow us to run the same software suite on hardware from different vendors and across processor generations.

It was also clear that Linux and Intel's Itanium systems are a very strong combination. We have been using them for quite some time and have seen very good performance on applications.

Binary Freedom: Is the work carried out going to be implemented by NCSA's own scientists, or do you see other commercial entities hiring time on these systems... if so, what do you think this change to Linux will mean for them? Did any of your potential commercial users express views on whether this attitude change was a good or bad idea?

Dan: NCSA is a National Science Foundation-funded supercomputing center, and as such, it provides computing resources to scientists across the country- not just at the center. In addition, NCSA has a number of partners in the commercial sector. These partnerships give the businesses access to leading-edge systems and provide us with feedback and information on where commercial computing is going. There is a lot of interest in the new clusters from NCSA partners in the private sector, as this is technology that they are very interested in evaluating. We have, in fact, worked with one of these partners very recently to help them set up their own Linux cluster at their site.

Binary Freedom: Let's get down to hardware. We gather from your press release that there are two clusters being built. Can you give us some specs on this?

Dave: NCSA's Linux clusters will include more than 600 IBM eServer xSeries Intel processor-based servers, running Linux and Myricom's Myrinet cluster interconnect network. Specifically, NCSA has purchased 512 IA32 xSeries servers and 160 Itanium eServers. The Itanium cluster, to be installed this summer, will be one of the first to use Intel's next generation 64-bit Itanium processor. The two clusters will expand the proven capability that NCSA has already demonstrated with Linux clusters and both Intel architectures.

Binary Freedom: IBM's had an interest in Linux for some time, but they haven't been the most enthusiastic of companies when it comes to a Linux strategy. Does IBM have the ability to manage a multiple-teraflop linux cluster, or will you be seeking support from Red Hat or Turbolinux?

Dave: We are working with our partners on a variety of initiatives, including clusters.

Binary Freedom: This is for Dan. High-performance computing's always been in the realm of companies like SGI and IBM with their proprietary architectures... S/390s and such. Personally, I've always considered x86 to be a cheap and mainstream platform designed for getting small and cheap parts on motherboards. Why the architecture move? Do you consider it a trade-off in performance against running, say, Linux on S/390 or an SGI Itanium system?

Dan: Many of the computing requirements for commercial computing are very similar to those for scientific computing: reliable operating systems on machines with fast processors as part of a highly scalable system. The second part of this purchase is for a large Linux Itanium cluster from IBM with 160 dual processor Itanium nodes, which illustrates exactly this point.

Binary Freedom: Linux is a system that's always been heavily based on Open Source code. Do you foresee many changes to the kernel, and have there been any decisions about whether or not to release these back into the public domain? Can we expect to see an NCSA-IBM distribution of Linux optimized for computation? Will this project affect "normal" Linux users in any way?

Dave: IBM is working closely with the open source community to build a better Linux and make the kernel more scalable. We are working on developments for the 2.6 kernel.

Thanks to Dave and Dan for their input. The scalability of Linux is often said to be over-hyped, but this news is clearly another step in Linux' recognition as a computational powerhouse as well as a competitive desktop OS.