DHPC Group's Beowulf Cluster Projects

Beowulf clusters are high-performance computers built from off-the-shelf commodity components. They usually consist of a cluster of PCs running Linux and connected using a Fast Ethernet network, however some clusters use high-end Unix workstations (such as Compaq Alpha or Sun UltraSPARC machines) and/or high-end gigabit networks (such as Myrinet, ServerNet or Giganet).

Beowulfs have become very popular over the past couple of years, due to the rapid improvements in the performance of commodity processors and networking infrastructure, and the development of Linux, a free Unix-like operating system for PCs. For most applications, clusters offer much better price/performance than standard supercomputers such as vector or shared memory machines.

We have been building and experimenting with Beowulf systems for two reasons: to construct cost-effective high-performance computers for use by computational scientists; and to explore some of the many research issues in parallel and cluster computing with Beowulf systems.

The DHPC group has designed and installed two large clusters which are among the most powerful supercomputers in Australia.

Beowulf Clusters

• Iofor was our first Beowulf system, built very cheaply in early 1998 using second-hand 486 PCs. It was used for experiments, software development and testing, and research into parallel I/O and data tiling.

• Perseus is our first purpose-built production system, developed for computational chemistry applications. Officially known as the South Australian Computational Chemistry Facility, the machine consists of 116 dual-Pentium PCs (i.e. 232 processors) connected by 100 Mbit/s switched Ethernet, with a peak speed of 113 GFlops. When commissioned in March 2000 it was the largest and fastest cluster in Australia and one of the largest PC clusters in the world.

• Orion is a commercial cluster from Sun Microsystems, used by physicists studying the fundamental forces of nature and the basic structure of matter (quarks, gluons, leptons, etc). Officially known as the National Computing Facility for Lattice Gauge Theory, the cluster is a Sun Technical Compute Farm consisting of 40 4-way E420R nodes (i.e. 160 processors) connected by a high-speed Myrinet network. It was the fastest computer in Australia when installed in June 2000, and ranked #188 in the November 2000 list of the Top 500 supercomputers in the world.

• MacBeowulf is a cluster of iMacs that normally run MacOS. By day it is a mild-mannered teaching laboratory, but at night or during holidays it can be transformed into a LinuxPPC-powered Beowulf super-cluster.

Research Areas

• Performance modeling and benchmarking of parallel programs
• Parallel I/O and data tiling
• Job scheduling, fault tolerance, and dynamic problem decomposition
• System design for particular applications, including scalability, performance and price/performance analysis
• Feasibility study of cluster computing on iMac clusters
• Cluster management

Technical Reports

1. DHPC-061: Beowulf - A New Hope for Parallel Computing?, K.A. Hawick, D.A. Grove and F.A. Vaughan, January 1999.
2. DHPC-065: Cluster Computing with iMacs and Power Macintoshes, D.A. Grove, P.D. Coddington, K.A. Hawick and F.A. Vaughan, March 1999.
3. DHPC-073: Commodity Cluster Computing for Computational Chemistry, K.A. Hawick, D.A. Grove, P.D. Coddington, and M.A. Buntine, January 2000.
4. Cluster Computing R&D in Australia, M. Baker, R. Buyya, K. Hawick, H. James and H. Jin, published as part of the Asian Technology Information Program (ATIP), Japan/USA, April, 2000.
5. DHPC-104: Precise MPI Performance Measurement Using MPIBench, D.A. Grove and P.D. Coddington, May 2001.

Beowulf Links

• Beowulf Project
• Beowulf Consortium - Beowulf Clusters Worldwide
• Extreme Linux
• Beowulf translated text (local copy)
• List of other relevant links

For more information, contact Paul Coddington.