run time

Discussion about analysis, visualization, and collaboration tools and techniques

Moderators: arango, robertson

Post Reply
Message
Author
pooran
Posts: 72
Joined: Fri Jan 31, 2014 5:54 pm
Location: Khoramshahr Marine Science and Technology Universi

run time

#1 Unread post by pooran »

Dear
i expect that by increasing number of nodes, model run time decrease linearly. For example: by choosing NTilei=1 and NTilej=1 and then Ntilei=2 and Ntilej=2 , run time decrease to 1/4 times. But it decreases just 1/2 times :?: .
Would you please tell me that if there are some things wrong in mpi program or run time is depends to some thing else :?: .
Cheers
pooran

moghimi
Posts: 18
Joined: Fri Jan 29, 2010 6:17 pm
Contact:

Re: run time

#2 Unread post by moghimi »

Hi Pooran,
Increasing number of nodes does not decrease computation time with the same ratio. There are other factors kicks in to prevent you from your desired results. Communication between nodes and load balancing is the most important ones.
Especially for ROMS, I suspect to load balancing the most, because you will not get the same number of active points in each computation core. You may get better efficiency if you use curvilinear grid to decrease your land areas.

Good luck,

Post Reply