How Grid Systems Pool Resources and Manage Tasks

How Grid Systems Pool Resources and Manage Tasks

by FaidTerence JABO -

Think of a Grid as a team of different computers (called "nodes") that pool their resources together to act like one giant, super-powerful machine.


1. What do they share?

  • Brainpower (CPU): This is the most common resource. You can use the grid to run a program on a faster machine, split a big task into tiny pieces to run all at once (parallel processing), or run the same task many times on different machines.

  • Storage: The grid combines the hard drives of many computers so they look like one massive library. This makes it easier to handle huge files, like data from scientific experiments.

  • Internet & Speed: Grids need fast "pipes" (bandwidth) to move data between machines.

  • Expensive Software: Instead of buying a license for every single computer, an organization can put the software on a few grid machines and send specific jobs to them to save money.

2. How is the work managed?

A Job is a specific task you want the grid to do. Since there are many machines, the grid uses a Scheduler (like a manager) to decide which computer should handle which task. There are three main ways machines participate:

  • Scavenging: The grid only uses a computer when its owner isn't using it (like when a clerk is at lunch). If the owner comes back, the grid task pauses.

  • Dedicated: The computer does nothing but grid work. This is faster and more predictable.

  • Reservation: You "book" a computer in advance for a specific time, like reserving a conference room.

3. Types of Grids

  • The Cluster: A small group of identical computers in one room.

  • The Intragrid: Different types of computers spread across different departments in the same company.

  • The Intergrid: A massive network that connects different companies or even countries. This is the most complex and requires the highest security.

4. Why use it?

The goal is Scalability. In a perfect world, if you double the number of processors, the work should get done in half the time. While "perfect" scaling is hard to hit because computers have to talk to each other, grids make impossible tasks possible by sharing the load.