Code Slows Down if I Preallocate Larger Variables?

Hello, I am using Matlab to explicitly integrate equations of motion for a finite element mesh. In the code I initialise all the history variables (such as displacement, velocity and acceleration) to store all the values at every node for every time step.

displace = zeros(gdl,n_inc+1); %
epsilon = zeros(gdl,n_inc+1); %
stress = zeros(gdl,n_inc+1); %

The point is: if I use a larger number of time steps (i.e. a smaller time increment), every single step takes much more to run. Is that avoidable somehow?

[Guillaume]

if I use a larger number of time steps the same simulation takes much more to run.

Why would you expect different? repeat something more times == do it for longer.

However, I don't understand how it relates to the title of your question.

[Alessandro]

I would not expect different run times, although I am experiencing it. I would think this is related to the initialisation of much larger matrices (which have DOF x n_increments dimensions). As an example, the exact same mesh takes 0.432s per step for a 5000 increments simulation, 0.712s for a 10000 increments one.

[Adam]

Why wouldn't you expect different run times? If you halve the time step you double the number of values over which to calculate which will slow your code down by somewhere up to the order of a factor of 2, dependent on what the code is and how dependent its runtime is on that setting.

[Alessandro]

I wouldn't expect different run times per step, since the number of required calculations per step is the same and I am not even close to saturating physical memory of the machine. In other words, if 100 steps take 1 minute, 200 steps shouldn't take much more than 2 minutes.

[Adam]

That depends on how the algorithm is coded. The profiler should tell you where the time is being spent.

profile on

...your code here...

profile off
profile viewer

[Walter Roberson]

32 bit MATLAB or 64 bit MATLAB?

[Alessandro]

64bit version.