News:

Building a 3D Ray Tracer  By stevmjon

Main Menu

Parallel computing

Started by apdf_2, April 10, 2005, 10:44:17 AM

Previous topic - Next topic

apdf_2

I am doing a project on Parallel Computing wherein several nodes(PCs in my case) are used to parallely process a huge task, therefore reducing the execution time for that task..I have sucessfully run a program on addition of squares of first 10000000 natural numbers, but the execution time did not decrease coz the time used in communication b/w nodes exceeded the advantage we cud get from parallel execution..can anyone please send me the code for a huge task that has parallel modules in it..Kindly use C language only and MPI programming Interface for the same..

Details about the working of my project:
LAM environment is first enabled in Linux and then MPI (Message Passing Interface ) is used to distribute parallel modules of the huge task to various nodes in the cluster...these nodes then execute their parts and return the results to the requesting node, also known as the master..MPI has predefined libraries which can be used to communicate bertween several processors..for more details, goto http://www.lam-mpi.org.

regards..
Puneet.

kevin

Sounds interesting, but since it's hardly a code example, I'll leave it here in chat.

RavenVegetable

Quotecan anyone please send me the code for a huge task that has parallel modules in it...Kindly use C language only and MPI programming Interface for the same..

No problemo, just send me the money. Kindly use U.S. currency only and crisp hundred-dollar bills for the same.

In the meantime,

-Double check that you're not misusing the API.
-Upgrade the network to something more suitable like gigether or fiber.
-Remember that in a homogenous cluster, the slowest bus governs the maximum throughput. I'm guessing yours is mix-and-match, so it's closer to an average of the slowest bus for each node. In either case, actual performance figures go even further down from there, after all overhead is factored in.
-Ensure the code responsible for task division is not assigning underweight tasks (where the processing time is less than the transportation and signaling times) especially in a high latency system.