Print

[Joseph]

I wrote a program and tried testing to see if it calculated faster by using my single computer as a grid. However, when I ran it with the grid, it takes a bit longer than when I run the program normally. When I think about it, the grid takes the job, splits it, sends it to nodes, and finally receives the tasks back from the nodes after they are done. What makes grid computing actually faster than just running it using a single processor? Is my job too small for there to be an obvious difference?

[lolo]

Hello,

Yes, you are right, there are some categories of jobs / workloads that will not benefit from being executed on a grid. The most trivial example is a single job with a single task. Since a task is the atomic workload unit for JPPF, it has nothing to parallelize here.

Generally speaking, a job is not worth running on a grid whenever the grid overhead outweighs any potential performance gain. This can happen for instance with jobs that have a small number of short-lived tasks. You may also have that issue with a very congested grid: when there is more work than there are nodes to perform it, then jobs may wait a long time in the server queue.

Without knowing anything about jour jobs and tasks, I cannot tell which caase you are in. I just hope my comments will spark some ideas and inspiration.

Sincerely,
-Laurent