This paper presents gain scheduled routing, a new scheme for reward-maximization in multi-service loss networks. Per-link control policies and their expected future rewards are precomputed using Markov decision theory and stored in lookup tables. Selection of appropriate table entries is based on measured or predicted call arrival rates. An approximation is introduced, which makes it possible to use identical lookup tables on links having the same capacities, although the links carry calls with different link reward parameters. Simulations with Poisson arrival processes show that gain scheduled routing offers significant improvement of the average reward rate, compared to least loaded path routing. Promising results are also obtained in simulations with self-similar call traffic.
Available as compressed Postscript (401 kB, no cover)
Download BibTeX entry.