Transportation applications, in particular traffic equilibrium applications,
are used by traffic engineers for traffic planning. We have
conducted extensive experiments to identify the inefficients of parallel
traffic equilibrium applications executed on a distributed memory machine.
Our analysis identified that the processor idle time is the major inefficiency
in parallel shortest path computations, the computationally-intense step
of traffic equilibrium problems. This analysis lead to the development of
efficient methods to significantly reduce this idle time resulting in
a significant reduction in parallel execution time. Further, we identified
the network parameters that are highly correlated with parallel computation
time. The results of this work can be used with decomposition
methods to produce load-balanced partitions.
This project is a collaboration with the Urban Transportation Center
at University of Illinois at Chicago.
