Grid computing is a distributed system made up of many connected machines (nodes) that share resources such as processors, storage, communication bandwidth, software, and special equipment. These machines may be different in architecture, operating systems, and capacities, but they work together as one system.
The most important resource in a grid is computing power (CPU cycles). It can be used to:
- Run an existing application on a remote machine.
- Divide an application into smaller parts that run in parallel.
- Run the same application many times on different machines.
Scalability measures how efficiently multiple processors reduce execution time.
Another key resource is data storage. When storage from different machines is combined, it forms a data grid. Storage can be memory (fast but temporary) or secondary storage (permanent). Grid file systems increase storage capacity, improve performance through data striping, provide a single namespace, and improve reliability using journaling and synchronization.
Communication is also essential. Machines must exchange jobs and data within the grid and sometimes access external networks like the Internet. Bandwidth and reliable connections are important to avoid bottlenecks.
The grid may also manage software and licenses, ensuring that jobs are sent only to machines where required software is installed and license limits are respected.
Machines in a grid may have different architectures, equipment, and policies. Schedulers assign jobs based on these attributes and organizational rules.
In grid terminology:
- An application is the overall task.
- It is divided into jobs.
- Jobs may be further divided into subjobs.
Some jobs can run in parallel, while others depend on the completion of previous jobs.
Scheduling is the process of assigning jobs to appropriate machines.
Scavenging allows idle machines to execute grid jobs.
Reservation allows resources to be booked in advance to guarantee performance.
Grids can exist within one organization (intragrid) or between multiple organizations (intergrid), with security becoming more important as the grid expands.
Overall, grid computing is about sharing distributed resources efficiently to perform large-scale and complex tasks.