Frontiers2002
photo

More than 50 Argonne scientists pitched in for the high-tech Chiba City barn raising.

Click on image for larger view.


‘Chiba City’ is a Mecca for computer scientists

"Chiba City" is the new destination for scientists developing the high-performance computer software of tomorrow. Named for the futuristic Tokyo suburb in William Gibson’s science fiction novel Neuromancer, this Chiba City doesn’t appear on any map of Japan — it’s a cluster of 256 Linux nodes at Argonne.

Linux computers have taken the advanced computing world by storm because they offer high performance at far less cost than traditional supercomputers. Large numbers of relatively cheap desktop-type or server computers can be connected to form a powerful supercomputer.

The big unknown about these cluster computers is scalability. Will existing numerical libraries scale to hundreds or thousands of nodes? What types of software and systems management tools can be built to make these large systems operate efficiently?

That’s where Chiba City comes in. System developers can run scalability tests, evaluating such factors as protocol performance and job scheduling on multiple nodes. Computer scientists can use the system for exploring new algorithms, communication libraries and parallel file systems. And computational scientists can compare Chiba City performance with that of Argonne’s other supercomputers.


BARN RAISING
Chiba City comprises 256 dual-CPU computational servers from VA Linux Systems. IBM Netfinity servers perform cluster management, file storage and visualization. These systems are divided into "towns" that carry out various functions from the actual computing to rendering images for visualizing the data.

The cluster installation was accomplished in a two-day "barn-raising" event. More than 50 Argonne scientists pitched in to help build the machine, which first opened to users in February 2000.

Examples of new technologies developed using Chiba City include:

  • The Parallel Virtual File System, or PVFS. Parallel file systems are critical for high-performance input/output on large clusters. PVFS is intended both as a high-performance parallel file system, which anyone can download and use, and as a tool for pursuing further research in parallel input/output and parallel file systems for Linux clusters.
  • The Multi-Purpose Daemon, or MPD. For interactive parallel jobs using message passing, this process management system has reduced startup time from minutes to seconds.
  • The first computations of 10-body nuclei. Argonne physicists developed methods for modeling the properties of atomic nuclei with up to 10 protons and neutrons. Nuclei contain complicated forces that are strongly dependent on the spins and charge states of the nucleons. These calculations are important to verify nuclear models for the neutron-rich nuclei to be produced at the proposed Rare Isotope Accelerator.


COMPUTER COMMUNITY

One goal of the Chiba City project is to build a community of open software developers and researchers. "Scientists in a community can learn from each other, build on shared experiences and develop a sense of group accomplishment," said Remy Evard, manager of advanced computing technologies and networking.

For more information please contact Dave Jacqué

Next: Did molecular hitchhikers ride a comet to Earth?

Frontiers2002 home page | About Argonne | Scientific Excellence
Research Facilities | Energy and Environment | Partners in Progress

Frontiers Archives – past issues | Contact the Editor | Argonne National Laboratory home page