processing time
 

Dear All,

Could anyone advise how or point me to a good reference for estimating processing time for a simulation?

We're currently running steady state turbomachinery simulations with CFX but are considering buying time on an external HPC facility. Unfortunately I have no idea how to predict processing time on either a desktop or parallel cluster so justification is tricky!

Any suggestions greatly appreciated.

Ianto

Re: processing time
 

Hi Ianto,

My personal formula is: number_of_cells * number_of_time_steps * iterations_per_timestep * number_of_equations / number_of_processors / a_constant = calculation_time_in_seconds

It does not work flawlessly, but it gives a rough estimation. When I started doing CFD the value for a_constant was about 50000, but it has gone up to about 250000 now.

Keep in mind that you cannot throw an infinite number of processors at a problem. This is especially true for time dependent simulations with a high number of time steps and a relatively modest grid.

Kind Regards, Wiebe.