Beowulf is the legendary hero from the Old English poem of the same name. In this epic tale, Beowulf saves the Danish kingdom of Hrothgar from two man-eating monsters and a dragon by slaying each.

Beowulf is used today as a metaphor for a new strategy in high-performance computing that exploits mass-market technologies to overcome the oppressive costs in time and money of supercomputing, thus freeing people to devote themselves to their respective disciplines. Ironically, building a Beowulf cluster is so much fun that scientists and researchers eagerly roll up their sleeves to undertake the many tedious tasks involved—at least for the first time they build one.

The concept of Beowulf clusters originated at the Center of Excellence in Space Data and Information Sciences (CESDIS), located at the NASA Goddard Space Flight Center in Maryland. The goal in building a Beowulf cluster was to create a cost-effective parallel computing system from mass-market commodity, off-the-shelf components to satisfy specific computational requirements in the earth and space sciences community.

Clusters of commodity computing systems

The first Beowulf cluster was built from 16 Intel DX4 processors connected by a channel-bonded 10 Mbps Ethernet, and it ran Linux. It was an instant success, demonstrating the concept of using a commodity cluster as an alternative choice for high-performance computing (HPC). After the success of the first Beowulf cluster, several more were built by CESDIS using several generations and families of processors and network interconnects.

At Supercomputing ’96, a supercomputing conference sponsored by the Association for Computing Machinery (ACM) and the Institute of Electrical and Electronics Engineers (IEEE), both NASA and the US Department of Energy demonstrated clusters costing less than $50,000 that achieved greater than a gigaflop-per-second sustained performance. As of November 1999, the fastest Beowulf-class cluster, Cplant (Computational Plant) at Sandia National Laboratory, ranked 44th in the Top 500 supercomputer list.

While many commercial supercomputers use the same processor, memory, and controller chips that are employed by Beowulf clusters, they also integrate proprietary gluing technologies (for example, interconnection networks, special I/O subsystems, and advanced compiler technologies) that greatly increase cost and development time. On the other hand, Beowulf clusters only use mass-market components and are not subject to delays and costs from custom parts and proprietary design.

Logical view of a Beowulf cluster

A Beowulf cluster uses a multicomputer architecture. It features a parallel computing system that usually consists of one or more master nodes and one or more compute nodes, or cluster nodes, interconnected via widely available network interconnects. All of the nodes in a typical Beowulf cluster are commodity systems—PCs , workstations, or servers—running commodity software such as Linux.

The master node acts as a server for Network File System (NFS) and as a gateway to the outside world. As an NFS server, the master node provides user file space and other common system software to the compute nodes via NFS. As a gateway, the master node allows users to gain access through it to the compute nodes. Usually, the master node is the only machine that is also connected to the outside world using a second network interface card (NIC).

The sole task of the compute nodes is to execute parallel jobs. In most cases, therefore, the compute nodes do not have keyboards, mice, video cards, or monitors. All access to the client nodes is provided via remote connections from the master node. Because compute nodes do not need to access machines outside the cluster, nor do machines outside the cluster need to access compute nodes directly, compute nodes commonly use private IP addresses, such as the 10.0.0.0./8 or 192. 168.0.0/16 address ranges.

From a user’s perspective, a Beowulf cluster appears as a Massively Parallel Processor (MPP) system. The most common methods of using the system are to access the master node either directly or through telnet or remote login from personal workstations. Once on the master node, users can prepare and compile their parallel applications, and also spawn jobs on a desired number of compute nodes in the cluster.

Applications must be written in parallel style and use the message-passing programming model. Jobs of a parallel application are spawned on compute nodes, which work collaboratively until finishing the application. During the execution, compute nodes use standard message-passing middle-ware, such as Message Passing Interface (MPI) and Parallel Virtual Machine (PVM), to exchange information.