Features:

INTERVIEW WITH PETER BRAAM, CTO CLUSTER FILE SYSTEMS
by Tim Curns, Assistant Editor

HPCwire: You've announced the first commercial release of the Lustre file system on December 15th. How long has it been in development?

BRAAM: We began as early as 2000 working with Seagate, which was looking for an object-oriented file system for enterprise use. But early we were approached by very high performance computing customers, such as Sandia and Lawrence Livermore National Labs, who were very excited because they felt the technology could scale well beyond 10,000 nodes. We made a decision to focus on highly scalable supercomputing environments as a proving ground, embracing an Open Source approach and providing pre-release versions to key customers. Earlier this year, LLNL felt the system was ready to take into general production use. They have played a major role, giving advice and feedback, and setting requirements leading up to this release.

HPCwire: What makes you think it is ready now?

BRAAM: With the help of our early users, the technology has been validated on an enormous scale in production use. In the just released top500 list Lustre is running on 4 of the top 10 supercomputers in the world, including 4 of the top 5 Linux-based supercomputers. It reached the fastest sustained data bandwidth in the world for Linux cluster applications at 11.1 GB/s parallel throughput. Perhaps even more importantly, Lustre has never lost customer data. So it is very definitely ready for prime time.

HPCwire: Who's going to need this?

BRAAM: The first customers are largely government labs, with applications like nuclear weapons testing, bio-informatics, and oil and gas exploration. But quickly it will move into other more commercial settings such as video and broadcast, and large enterprises trying to manage distributed data. Lustre will give an administrator complete control over storage clusters, allowing them to quickly add or manage storage at a very low cost.

HPCwire: What makes it any different from other high-end storage systems like GPFS or PVS?

BRAAM: The key advantages are cost, scalability, and ease of management. Lustre can scale more than ten times the capacity of previous systems for about one tenth of the cost per terabyte. It's also much more manageable than existing large-scale storage systems. For example, you will be able to add storage server on the fly, at any location, and data will be distributed dynamically across the enterprise.

HPCwire: With the release, will your focus now become enterprise applications?

BRAAM: Our first priority is to bring as many high performance computing environments into production as possible because they will push the technology to its limits and give us a well-tested and mature feature set. We believe the enterprise storage companies will quickly have an interest, however, and we expect Lustre to move into broad-based usage within the next couple of years.

HPCwire: Will this be the end of your Open Source approach?

BRAAM: No, Lustre will continue to be developed and maintained as Open Source source software. Like the Open Source operating system, Linux, the Lustre file system should be available everywhere and free to those who choose that path. But we will also move to a much more defined, paid release cycle after 1.0. Customers that want support, extensions and well-tested enhancements will undoubtedly feel more comfortable with the paid version.

HPCwire: Where will I get it?

BRAAM: Cluster File Systems, Inc. will continue to work directly with customers to support and install the system with its release in December, but we are already beginning to develop partnerships with other vendors who will be integrated Lustre into their storage offerings . And of course Open Source users will still be able to download it free from our web site.