Hi all.
I'm running ROMS on a Linux box with two Intel Xeon processors (3.2 GHz). So far I've successfully run ROMS using mpi. The dimension of my grid is 240x130x30 and when I use all 4 processors the computational time per time step is about 4 seconds.
My question regards the amount of RAM I'm using to run te model. My machine has a total of 1 Gb of RAM. Even if it seems to be enough for my simulation (even if all the RAM is used, the machine doesn't swap while running ROMS...that's what shows up using the Linux command "top"), I was wondering if it might still limit the performances of the machine, and therefore if increasing the amount of RAM (I can upgrade up to 8 Gb) might reduce the amount of time required to run the model.
Thank you,
Francesco
How much RAM??
Re: How much RAM??
Will you ever want to run a bigger problem or turn on more tracers? Will you ever want to run with the DIAGNOSTICS output? What sort of post-processing/visualization do you expect to do? Any of those can easily bump you over that 1 GB.
-
nencio
Re: How much RAM??
Hi kate,
thanks for replying.
Well, I've already run the model with the DIAGNOSTIC activated and still I haven't noticed any swapping during the simulation. Regarding the number of tracers and grid domain: for the first part of my experiment I won't run with more than T and S, and the grid will be the one I'm using now.
To post-process and visualize the data I'm using Matlab, and so far I haven't had too many problems with plotting the data (even though I cannot load the output netcdf as a whole, but I have always to go a timestep or a slab at the time....).
I run at first with only one processor and I've only recently started using mpi. This allowed me to reduce the computational time by a half. Now I'd like to know if increasing the amount of RAM might further improve the computation speed, or if I've already reached the limits of my machine.
Thank you again,
Francesco
thanks for replying.
Well, I've already run the model with the DIAGNOSTIC activated and still I haven't noticed any swapping during the simulation. Regarding the number of tracers and grid domain: for the first part of my experiment I won't run with more than T and S, and the grid will be the one I'm using now.
To post-process and visualize the data I'm using Matlab, and so far I haven't had too many problems with plotting the data (even though I cannot load the output netcdf as a whole, but I have always to go a timestep or a slab at the time....).
I run at first with only one processor and I've only recently started using mpi. This allowed me to reduce the computational time by a half. Now I'd like to know if increasing the amount of RAM might further improve the computation speed, or if I've already reached the limits of my machine.
Thank you again,
Francesco
Re: How much RAM??
If you aren't swapping now, then more memory isn't going to help. The size of your cache is more likely to be limiting.