Grid computing

Grid computing is a method of using widely distributed computer resources to reach a common goal. It functions as a distributed system that manages non-interactive workloads involving many files. This technology allows many networked, loosely coupled computers to act together to perform large tasks. While a single grid might be dedicated to one specific application, they are often used for many different purposes. A grid can be quite large in scale. It can range from a small network of workstations within one corporation to a massive public collaboration involving many different companies and networks.

To understand how it works, we can look at the individual components. A grid relies on complete computers rather than just single processors. Each node in the grid has its own onboard CPU, storage, power supply, and network interface. These computers connect to a private or public network, such as the Internet, through a conventional network interface like Ethernet. This setup makes it a special type of parallel computing. In this model, multiple computations can take place independently. This is very useful for tasks that do not require processors to communicate intermediate results constantly.

Grid computing is distinct from other high-performance systems like cluster computing or traditional supercomputers. In cluster computing, nodes are often physically coupled and perform similar tasks. In contrast, each node in a grid is often set to perform a different task or application. Grids are also more heterogeneous, meaning they use a variety of different hardware and software. A traditional supercomputer uses many processors connected by a local, high-speed computer bus. A grid uses commodity hardware connected over a standard network. This means grids generally have lower efficiency because the processors and local storage areas lack high-speed connections.

Managing these systems requires specialized software called grid middleware. Middleware acts as a layer between the hardware and the software. It allows different, heterogeneous resources to be shared across multiple administrative domains. This software helps manage complex technical areas like security, data management, and virtual organization management. Some major middleware projects include the Globus Toolkit, gLite, and UNICORE. Developers also use grid workflow systems to manage tasks. These systems are designed to compose and execute a specific series of computational or data manipulation steps.

Because grids often use computers from different owners, designers must handle issues of trust and availability. One major challenge is that participating nodes might not be entirely trustworthy. To prevent malicious participants from producing false or erroneous results, designers use specific measures. They might assign work randomly to different nodes and check if at least two nodes report the same answer. If the answers differ, it identifies a malfunctioning or malicious node. Additionally, nodes can drop out of the network at any time. Laptops or users on dial-up connections may become unavailable unpredictably.

To handle these uncertainties, grids use specific strategies like CPU scavenging. This technique exploits the idle resources in a network of participants. It harvests "spare" instruction cycles that occur when computers are inactive, such as at night or during lunch breaks. Many volunteer computing projects, such as BOINC, use this scavenging model. To manage nodes that go offline, the system can assign large work units. This reduces the need for constant network connectivity. If a node fails to report results in the expected time, the work unit can be reassigned.

Grid computing has significant applications across many professional fields. In science, it is applied to intensive mathematical and academic problems. Commercial enterprises use grids for drug discovery and economic forecasting. They also use them for seismic analysis and back office data processing for e-commerce. The market for this technology is divided into several segments. The provider side includes the middleware market, grid-enabled applications, utility computing, and Software as a Service (SaaS). Utility computing provides grid resources as a service, with major players including IBM, HP, and Sun Microsystems.