What is grid computing?

Grid computing refers for a cluster of de­cen­tral­ized computers that form a virtual su­per­com­put­er. The flexibly dis­trib­uted computing power makes it possible to perform complex tasks with multiple resources si­mul­ta­ne­ous­ly and to optimize in­fra­struc­ture uti­liza­tion.

Grid computing is a sub-area of dis­trib­uted computing, which is a generic term for digital in­fra­struc­tures con­sist­ing of au­tonomous computers linked in a computer network. The computer network is usually hardware-in­de­pen­dent. This means that computers with different per­for­mance levels and equipment can be in­te­grat­ed into the network. Dis­trib­uted ap­pli­ca­tions and processes can work across devices with networked computer units. The computer units, in turn, can com­mu­ni­cate with each other locally and across regions within the network and solve problems.

The dis­tinc­tion between dis­trib­uted computing and grid computing is fluid. Dis­trib­uted computing can refer to de­cen­tral­ized data pro­cess­ing in computer networks. Grid computing, on the other hand, refers to a virtual su­per­com­put­er that is created by con­nect­ing loosely coupled computers. This is used to handle com­pu­ta­tion­al­ly intensive processes or tasks. Linked servers and computers make their resources and computing power available to scale up to a required computer per­for­mance.

In grid computing, the strengths of computer clusters are not cen­tral­ized and are used supra-re­gion­al­ly in the form of grids. While computer clusters usually consist of locally limited computer networks, grid computing accesses computer ca­pac­i­ties within a computer network on a supra-regional basis. Not only computers are networked, but also databases, hardware, software, and computing ca­pac­i­ties. Within the framework of the grid, providers link globally and locally dis­trib­uted computer resources via in­ter­faces (nodes) and mid­dle­ware. They then assign these to virtual or­ga­ni­za­tions, which in turn determine which resources can take over tasks or how computing power can be optimally dis­trib­uted for an ap­pli­ca­tion.

Grid computing is used both for com­mer­cial purposes and for sci­en­tif­ic and economic data analysis and pro­cess­ing. If complex processes exceed the computing power of a computer or a local computer cluster, grid computing can help to integrate, evaluate, or display large amounts of data. Special hardware is not a pre­req­ui­site for grid computing. Rather, mid­dle­ware (software for ex­chang­ing data between ap­pli­ca­tions) on coupled computers ensures that free computing capacity is available within the virtual or­ga­ni­za­tion.

Grid computing is not limited to specific ap­pli­ca­tion areas, as the in­ter­con­nec­tion of computer clusters can serve a wide variety of purposes. Well-known areas of ap­pli­ca­tion for virtual su­per­com­put­ers are sci­en­tif­ic and economic big data analyses that work with enormous amounts of data and com­pu­ta­tion­al­ly intensive sim­u­la­tions. This applies to research in the natural sciences and medicine, but also in me­te­o­rol­o­gy, the in­dus­tri­al sector, or particle physics. An example of this includes the large-scale ex­per­i­ments of the Large Hadron Collider, CERN.

To define and classify grid computing in com­par­i­son to other tech­nolo­gies like cluster computing or peer-to-peer computing, three main cor­ner­stones can help:

• De­cen­tral­ized, local, and global co­or­di­na­tion of resources such as computer clusters, data analytics, mass storage, and databases.
• Stan­dard­ized, open in­ter­faces (nodes) and mid­dle­ware (protocols or protocol bundles) that connect computing units to the main grid and dis­trib­ute tasks.
• Provision of non-trivial quality-of-service (QoS) to optimally dis­trib­ute data streams and ensure constant scal­a­bil­i­ty and reliable data transfer under high com­pu­ta­tion­al demands.

Beyond this, grid computing can be divided into different clas­si­fi­ca­tions:

• Computing grids: The most common form of grid computing, where grid users use the coupled computing power of a virtual su­per­com­put­er via grid providers to dis­trib­ute or scale com­pu­ta­tion­al­ly intensive computing processes.
• Data grids: Data grids provide the computing capacity of in­ter­con­nect­ed computers to evaluate, display, transmit, share, or analyze large amounts of data via grid nodes.
• Knowledge grids: This structure uses the su­per­com­put­ing ca­pa­bil­i­ties of the grid to scan, connect, collect, evaluate, or structure large data sets and knowledge bases.
• Resource grids: These systems define coupled hi­er­ar­chies of grid providers, grid users, and resource providers in the grid. A role model de­ter­mines which resource providers can provide storage and computing ca­pac­i­ties, data sets, software and hardware, ap­pli­ca­tions, sensors, measuring devices, and other in­stru­ments via in­ter­faces.
• Service grids: In the service grid, grid service providers make resource providers’ bundled com­po­nents and ca­pac­i­ties available to grid users as a complete service. This demon­strates that grid computing combines service ori­en­ta­tion and computing services.

• Co­or­di­na­tion and man­age­ment of cross-device processes and tasks.
• Cost-effective scaling of business processes through coupled computing power and storage ca­pac­i­ties.
• Si­mul­ta­ne­ous/parallel pro­cess­ing, analysis, and pre­sen­ta­tion of large amounts of data through global computer networks.
• Complex tasks can be solved faster and more ef­fec­tive­ly.
• Reliable uti­liza­tion and optimal use of IT in­fra­struc­ture through virtual or­ga­ni­za­tions and flexible task dis­tri­b­u­tion.
• Low sus­cep­ti­bil­i­ty to failure, as ca­pac­i­ties are dis­trib­uted flexibly and modularly in the grid.
• No need for large in­vest­ments in server in­fra­struc­ture.

• Complex ad­min­is­tra­tion and in­com­pat­i­ble system com­po­nents can occur.
• Computing power does not increase linearly with the number of coupled computers.