Stacey Quandt

The growing number of Linux supercomputers is evidence of a significant shift for both the academic and commercial communities. Twice each year since 1993, the TOP500 (www.top500.org) supercomputer organization has published a list of the 500 most powerful computer systems based on the LINPACK performance benchmark that measures performance among all classes of supercomputers. While the Hewlett-Packard Superdome and AlphaServer, IBM SP and Sun Microsystems 15K systems continue to dominate the Top500 supercomputing list, there is also an increase in the number of Linux clusters. The highest ranking Linux system, at No. 35 on the Top500 list, is the University of Heidelberg Linux Cluster System (HELICS) running 825 Gigaflops and is based on AMD Athlon.

The HELICS supercomputer performs scientific research in the fields of reactive flows, human genome decryption, bioinformatics and applied physics at the Interdisciplinary Center for Scientific Computing (ICSC). In addition, the Tokyo Institute of Technology Presto III Cluster located in Tokyo, Japan also runs on AMD Athlon and ranks at No. 47. This is the first time AMD processor-based systems have earned places in the top ten percent of the TOP500 rankings. Another implication of the Top500 rankings is that it demonstrates that a single university lab can achieve terascale performance (a teraflop is equivalent to one trillion floating point operations per second).

With a total of 42 Intel-based and 7 AMD-based PC clusters on the Top500 list, there are broader implications for the growing adoption of Linux clusters. For example, IBM accounts for the most Beowulf supercomputers with 31 out of a total of 49 on the Top500 list. Dell also has a system in the top 100 with a 256 Intel Xeon system deployed at Sandia National Laboratories, ranking at No. 79. Increasingly, users in vertical segments such as oil and gas, automotive, life sciences, research and financial services are adopting Linux supercomputers. The implication of Intel and AMD x86-based systems is that teraflop level Linux machines are becoming the norm and Linux commodity clusters are redefining high performance computing.

Users have more options for high-performance computing today to reduce the cycle of development and deployment costs. Giga recommends Linux clusters of commodity x86-based hardware due to the ability to shorten cycle times for biocomputing, reactive flows, optimization problems, technical simulations and image processing. To a large degree this is forcing proprietary systems running Unix into a high-end niche market.

For users considering Linux clusters, there are caveats to bear in mind; many applications do not lend themselves to parallelized commodity clusters (for example, mixed workloads or applications that demand inordinate memory and throughput). Linux clusters are a rapidly evolving market segment, with further developments in cluster-based algorithms, cluster hardware, system configuration and cluster management anticipated.