Computer cluster

A computer cluster is a collection of multiple computers working together. They are designed to function as if they were one single, powerful system. In this setup, each individual computer is referred to as a node.

Most clusters connect these nodes through a fast local area network. While a single computer has limits, a cluster can expand its power by adding more nodes. This makes clusters a vital tool for modern computing, ranging from small business setups to the world's fastest supercomputers.

The mechanism of a cluster relies on specialized software called clustering middleware. This software layer sits on top of the individual nodes. It allows users to interact with the entire group as a single cohesive unit, a concept known as a single system image. In a common design called a Beowulf cluster, the system uses a "Master" computer to manage everything.

The Master computer handles all scheduling and management tasks. It typically uses two network interfaces. One interface communicates with the general organization network. The other interface connects to a private network used only by the "slave" computers. These slave nodes have their own local memory and disk space, but they can also access a large, shared file server for global data.

Clusters are often configured for specific types of work. One common method is load balancing. In a load-balancing cluster, the nodes share the computational workload to improve overall performance. For example, a web server cluster might assign different user queries to different nodes. This optimizes the response time for everyone. Another method is high-availability, also called a failover cluster. These systems use redundant nodes to ensure the service stays online. If one component fails, the redundant nodes take over. This provides fault tolerance, which is the ability to keep operating despite a malfunctioning node.

The history of clustering is closely tied to the evolution of computer networks. Some experts, such as Greg Pfister, suggest that customers began creating clusters in the 1960s. They did this because they needed more power or a backup for their work. A major milestone occurred in 1967 when Gene Amdahl published a seminal paper on parallel processing called Amdahl's Law. The first production system designed specifically as a cluster was the Burroughs B5700 in the mid-1960s. This system allowed up to four computers to connect to a common disk storage subsystem. This allowed them to distribute their workload efficiently.

Commercial clustering grew significantly in the following decades. In 1977, Datapoint Corporation developed the "Attached Resource Computer" (ARC) system. This was the first commercial loosely coupled clustering product. Later, Digital Equipment Corporation released the VAXcluster in 1984. This product was very important because it supported parallel computing, shared file systems, and peripheral devices. Other notable early systems include the Tandem NonStop from 1976 and the IBM S/390 Parallel Sysplex from around 1994.

While early supercomputers like the Cray 1 relied on internal parallelism, many modern supercomputers now use cluster architectures.

The significance of clusters lies in their scalability and cost-effectiveness. Clusters allow for horizontal scaling. This means you can improve performance and reliability simply by adding more computers to the group. This is often much cheaper than buying one massive, expensive computer with the same speed.

Clusters also allow for easier maintenance. A single node can be taken offline for repairs while the rest of the cluster continues to handle the workload. This prevents the entire system from having to shut down. This makes them ideal for high-performance tasks like weather simulations or modeling vehicle crashes.

Today, clusters are used in surprising ways. Because game consoles have become so powerful, some people repurpose them into high-performance computing clusters. Examples include clusters made from Sony PlayStation or Microsoft Xbox consoles. Additionally, high-end graphics cards are often used in clusters. These cards use a Graphics Processing Unit, or GPU, to perform calculations. Using GPUs can be much more economical than using standard central processing units, or CPUs, especially for certain types of computing tasks.