Estimating Runtimes

From Tuflow
Jump to navigation Jump to search

This page contains more information on estimating runtimes in TUFLOW. This only looks at the 2D computational time, for most cases the 1D calculations would add less than 1% to the runtime. Note the below is for TUFLOW "Classic" and not TUFLOW HPC ran either on GPU or CPU.

The spreadsheets linked below can be used to estimate TUFLOW simulation times based on the cell size, catchment area and simulation time (i.e. event duration). The runtime depends on a number of factors such as the speed of the CPU and also the fraction of the model that is wet. The more wet cells the more calculations that are required and therefore the slower the simulation.

The timestep used in a TUFLOW simulation is dependent on a variety of factors such as the steepness of the topography, the depth of water and the velocities encountered, in the spreadsheet a typical range of timesteps is calculated based on the cell size. For a TUFLOW model running in metric, the timestep (in seconds) is typically between 1/5 to 1/2 of the cell size (in metres). For example a 10m model will typically have a timestep between 2 and 5 seconds. Therefore, two estimates of runtime are provided in the spreadsheet one based on the smaller and one based on the larger timestep.

The second sheet in the attached excel files is titled Cell_Calcs_per_second and gives a guide to estimating this on your machine.

Given the inexact nature of the calculations these are estimates only, but do give a reasonable indication of runtime!

File:Estimate runtimes metric 001.xlsx
File:Estimate runtimes Imperial 001.xlsx