Implementing Computational Grid and Measure Its Throughput

Abstract

The motivating fact for my project is : “Distribute the job among several workers (Computers) to reduce the total processing time and to utilize their resources.”

High-performance distributed computing appears to be shifting away from tightly-connected supercomputers to “Computational Grids” composed of heterogeneous systems of networks, computers, storage devices, and various other devices that collectively act as a single geographically distributed “virtual” computer.

Nowadays there is a need for supercomputers on which people can run long compute jobs. To fulfill that need we have setup the Computational Grid. We have developed the networked system that acts as a Grid. One of the grid participants acts as the submission host from where we are going to submit the host. The main goal of computational grid is to reduce the processing time by distributing the long task among multiple machines which have their computation cycles free. We have achieved this by dividing the application amongst multiple parallel tasks and then distribute these tasks among the machines which are registered as the participants of the Grid and have their computational cycles free. Each individual machine has the task it has to do and all machines in the grid have the same algorithm to work on those tasks. For this the machines which are free and interested in participating in the grid register themselves to the Server with their available capacity. And they updates this information periodically so that the grid has the latest status of its nodes. According to the number of nodes available, the required job is divided into subjobs. These subjobs with the required data and other information are sent to the execution hosts. After processing its task, each execution host sends the response back to the submission host who collects all the outputs from execution hosts and present the final output to the submitting user in completely transparent manner.