contents

|
|
Efficient Algorithms for Path-Based and Dynamic Flow Problems in Large Networks
Abstract:
Efficient algorithms are of crucial importance for a variety of
network-based applications like traffic management and planning,
telecommunications, and data traffic in the Internet. A typical
example of such applications is the minimization of network load by
simultaniously satisfying various user constraints. This naturally
leads to optimization problems in networks, which are usually treated
with standard techniques from network optimization, such as algorithms
for shortest paths, maximum flows, min-cost multi-commodity flows. To
cope with the various user constraints that often relate to the
quality of the flow-carrying paths in the given solution, path-based
formulations for different flow problems are of interest.
While the classical techniques for solving flow problems are efficient
for common networks of smaller size and complexity, they often ignore
substantial aspects that are characteristic of larger, more complex
networks. Due to their input size, large and complex networks can
generally not be supplied explicitly but instead have to be given in
some implicit manner. For example, a traffic network can be modelled
via a hierarchical structure where different street classes (local
streets, highways, etc.) define different levels of hierarchy.
Unfortunately, standard algorithms for network optimization are
generally not capable of using this condensed representation of the
networks but require the full expanded network.
In this project we want to develop and test algorithms for solving
"path-based" flow and multi-commodity flow problems on large and
complex networks. The theoretical challenges here are
two-fold. First, the algorithms must be able to solve problems using
only partial information about the network structure, since it is
given implicitly. Second, the constraints of our problems
call for new path-based algorithmic approaches in order to solve flow
problems that are tailored for our large networks.
The practical test bed for these algorithms are traffic routing
problems, on which we cooperate with DaimlerChrysler. For these
applications, practical data is available for testing the
developed algorithms.
Jahres-Kolloquium_2002_talk_1on1.pdf
Jahres-Kolloquium_2002_talk_2on1.pdf
|