Computational Speed

Since in the application, many of the problems are fairly large, one needs to keep an eye on computing speed. A useful measure for this are ``vehicle updates per second''. Let's say that for a simulation with $10^4$ vehicles and $10^3$ time steps we need 10 seconds of computing time. Then we have $10^4 \times 10^3 = 10^7$ vehicle updates per 10 seconds, or $10^6$ vehicle updates per second. This number is typical for a simple implementation on a 300 MHz CPU.

Under unix one obtains the computing speed for example via time (see man-page). My personal result looks like

{}
92.88user 0.00system 1:34.50elapsed 98%CPU (0avg...

We are most interested in ``92.88user'' (coresponding to 92.88 sec).

Transportation science sometimes does the ``real time limit'' (for our purposes $=$ the number of vehicles with which the simulation runs as fast as reality).

All of these values depend on the vehicle density, which therefore always needs to be given when giving computing speeds.

Task 9.5   How long does your simulation for the ``corridor'' network with 0.plans take to run? Please also tell us your implementation (C++ or Java or ??). Do this once with output and once with output switched off. What does this roughly correspond to in ``vehicle updates per second''. How did you obtain that number?