Optimal network flow allocation

The main goal of this project is to investigate the network flow allocation (routing) problem, and devise a practical algorithm for its solution. Particularly, we plan to investigate the trade-off between the algorithm’s optimality and its implementability; therefore, obtaining an 2-suboptimal solution would be sufficient, if this will speed-up convergence or simplify the overall algorithm. This problem was brought to our attention during a meeting with Prof. McKeown and his Ph.D. student Yashar Ganjali. Initial project research will include a survey of the previous work performed by Yashar [Gan04]. In [GK98], several network flow allocation and fractional packing problems, including maximum multi-commodity flow, spreading metrics, maximum concurrent flow, and maximum cost-bounded concurrent flow, are solved using a dual problem formulation of their corresponding linear cost primal problems. First, we plan to examine the extension of the techniques described in the paper to general convex cost functions, such as the ones described in [Gan04]. Furthermore, we will look at different cost functions for optimizing performance, beginning with the classic cost function of average delay for M/M/1 queues,