Cloud Architecture

Types of Grid Computing and How Grids Differ From Clouds

The main types of grid computing, including volunteer grids, with examples such as the Worldwide LHC Computing Grid, and how grids differ from clouds.

On this page
  1. What is grid computing?
  2. Why there is no single list of grid types
  3. Types of grid by function
  4. Types of grid by scale
  5. Middleware and virtual organizations
  6. What is an example of grid computing?
  7. How grids differ from clouds
  8. Common misconceptions
  9. Where grids fit today

What is grid computing?

Grid computing connects computing resources from several organizations so they can work on one large problem. Unlike a single data center, a grid usually spans multiple administrative domains, which means several institutions each control their own hardware and policies. Resources are shared, and each participant keeps ownership and control of what it contributes.

Grid computing grew out of scientific computing. Researchers needed more computing power than any one institution could provide for large simulations, data analysis, and experiments. The Grid: Blueprint for a New Computing Infrastructure by Ian Foster and Carl Kesselman (1998) is a foundational text. In a 2002 checklist, Foster described a grid as a system that coordinates resources not under centralized control, uses standard, open, general-purpose protocols, and delivers a nontrivial quality of service.

Why there is no single list of grid types

Classifications of grid computing vary by source, and there is no single standard list. The categories below are common, but treat them as useful labels rather than a fixed taxonomy.

Types of grid by function

  • Computational grid: shares processing power to run large calculations, often as many independent jobs.
  • Data grid: provides distributed storage and access to large datasets, including replication and transfer between sites.
  • Collaborative grid: shares tools and workspaces among people working on the same project.
  • Scavenging or desktop grid: uses idle processor cycles on desktop computers.
  • Volunteer computing: a subtype of scavenging grid in which members of the public donate spare computing time. It is one kind of grid, not the whole category.

Some sources also describe network grids and utility grids. They are less commonly discussed.

Types of grid by scale

Grids can also be described by size. They range from:

  1. a cluster or departmental grid, inside one department
  2. an enterprise or campus grid, across one organization
  3. a global grid, across institutions and countries

Middleware and virtual organizations

Grids depend on middleware, the software that handles resource discovery, job scheduling, data movement, and security across organizations. Examples include the Globus Toolkit, HTCondor, gLite, UNICORE, and ARC. The Open Grid Services Architecture (OGSA) is a standards effort for grid services.

Users in a grid are often grouped into virtual organizations. These span institutions and define who may use which resources.

What is an example of grid computing?

  • Worldwide LHC Computing Grid (WLCG): a tiered grid that stores and processes data from the Large Hadron Collider. Tier 0 is at CERN, and the grid includes roughly 170 sites in more than 40 countries. The exact figures change over time.
  • Open Science Grid (OSG): a research grid in the United States. European Grid Infrastructure (EGI) serves a similar role in Europe.
  • Volunteer projects: SETI@home, which launched in 1999 and stopped issuing new work in March 2020; Folding@home, from 2000; and projects built on the BOINC platform, from 2002.

How grids differ from clouds

A cloud is owned and operated by one provider, which rents out virtualized resources on demand. A grid is shared by many organizations, and much of its hardware is owned by the participants. The table below summarizes the main differences.

Aspect Grid computing Cloud computing
Ownership Shared among participating institutions Owned and operated by one provider
Hardware Often varied, owned by participants Provider-owned pools of virtualized resources
Administration Spans multiple administrative domains Managed by a single provider
Access Through middleware and community agreements On demand, through consoles, CLIs, and APIs
Typical workloads Batch and high-throughput jobs Elastic services, web applications, and interactive workloads
Billing model Based on membership and resource sharing Pay-as-you-go metering

Grids favor batch and high-throughput jobs, while clouds favor elastic services and interactive workloads. Cloud platforms also rely heavily on virtualization, which is not a core feature of most grid systems.

Common misconceptions

  • A grid is not the same as a cluster. Clusters are co-located, use mostly identical nodes, and sit under one administrator. The comparison of clusters, grids, and clouds covers the differences in detail.
  • Grid computing is not obsolete. WLCG and OSG still process large scientific workloads.
  • Cloud is not simply a commercial grid. Cloud borrowed ideas from grids, but adds virtualization, self-service, and a utility billing model.

Where grids fit today

Grid computing remains important for scientific communities with large, shared, batch-oriented workloads. The problem of assigning jobs to shared machines is still central to both grids and clouds, and task scheduling explains the main approaches.

The complete guide to cloud architecture

Cloud Computing Architecture: The Core Components
Read the guide

Keep exploring

Guides from other topics

All topics →