Next: Summary
Up: No Title
Previous: Speed-up and efficiency
As explained in the introduction, a micro-simulation in a software
suite for transportation planning would have to be run many times
(``feedback iterations'') in order to achieve consistency between
modules. For the microsimulation alone, and assuming our
16 CPU-machine with switched 100 Mbit Ethernet, we would need about
30 hours of computing time in order to simulate 24 hours of traffic
fifty times in a row. In addition, we have the contributions from the
other modules (routing, activities generation). In the past, these
have never been a larger problem than the micro-simulation, for
several reasons:
- The algorithms of the other modules by themselves did
significantly less computation than the micro-simulation.
- Even when these algorithms start using considerable amounts of
computer time, they are ``trivially'' parallelizable by simply
distributing the households across CPUs.
- In addition, during the iterations we never replan more than
about 10% of the population, saving additional computer time.
In summary, the TRANSIMS modules besides the traffic micro-simulation
currently do not contribute significantly to the computational burden;
in consequence, the computational performance of the traffic
micro-simulation is a good indicator of the overall performance of the
simulation system.
Next: Summary
Up: No Title
Previous: Speed-up and efficiency
Thu Oct 5 16:59:00 CEST 2000