Biowulf -- Parallel Supercomputing on the "Cheap"

Biowulf—Parallel Supercomputing on the "Cheap"

The latest high performance computer system to join the Helix Systems is Biowulf, a 224-processor supercluster of PCs. That’s right, PCs—just like the one on your desktop. However 224 processors, running the Linux operating system and interconnected by a high-speed network, offer an effective and low-cost way to solve some very large computational problems in bioscience.

Biowulf reflects a recent trend in high performance computing—the use of a multiplicity of commodity computers to build a supercomputer. These "Beowulf"-class computers were first developed at NASA’s Goddard Space Flight Center, where the name Beowulf was given to the first machine. Using commercial "off-the-shelf" parts allows economies of scale that traditional high-performance computers cannot match. Although in many cases the absolute performance of Beowulf-clusters does not match SGI Origins, Cray T3s and IBM SPs, the price/performance ratios of clusters are 5-to-10 times better.

What’s in a Cluster

The Biowulf cluster consists of 112 dual-processor Pentium 450 MHz and 550 MHz nodes, each with 256 MB of memory and an eight GB disk, running the RedHat flavor of Linux. Each node connects to a fast Ethernet switch (100 Mb/s), over which node-to-node communication can occur using MPI (message passing interface), the standard communication library for parallel programs running on distributed memory machines. For applications that can take advantage of added capabilities, some nodes contain as much as 1 GB of memory and others are connected to a gigabit speed network. In addition, the Biowulf cluster has access to a high performance file server that shares the same file space as the other Helix Systems (Helix, Churn, Galaxy and Quasar).

Hardware Configuration

How Biowulf Is Being Used

Biowulf offers a full development environment, including high-performance compilers from The Portland Group, a software company specializing in compilers and development tools for Intel CPU architectures.

Since going production in October 1999, Biowulf has been running a wide range of scientific applications. Many are not even written as traditional parallel applications, but rather are run as "swarms" of 16 or more single-threaded jobs. Examples of applications running on Biowulf include:

DNA/protein sequence analysis
computational chemistry
linkage analysis
microarray analysis
statistical analysis

As an example of the potential benefit in using a parallel computer—running a ‘blast’ sequence analysis of 10,000 sequences against a human genome database (estimated at over 2 million sequences) on 64 Biowulf processors takes 50 minutes. Processing the same job on a typical desktop PC or workstation would take 2 or more days!

More Information

Visit the Biowulf home page at http://biowulf.nih.gov or contact the Helix Systems staff at staff@helix.nih.gov.

Interface 214 (June 23, 2000)

Other Issues and "Back Pages" | Comments