Two techniques for fast computation of constrained shortest paths

Computing constrained shortest paths is fundamental to some important network functions such as QoS routing, which is to find the cheapest path that satisfies certain constraints. In particular, finding the cheapest delay-constrained path is critical for real-time data flows such as voice calls. Because it is NP-complete, there has been much research into designing heuristic algorithms that solve the ε-approximation of the problem with an adjustable accuracy. A common approach is to discretize (i.e., scale and round) the link delay or link cost, which transforms the original problem to a simpler one solvable in polynomial time. The efficiency of the algorithms directly relates to the magnitude of the errors introduced during discretization. We propose two techniques that reduce the discretization errors, allowing faster algorithms to be designed. Reducing the overhead of the costly computation for constrained shortest paths is practically important for the design of a high-throughput QoS router, which is limited by both processing power and memory space. Our simulations show that the new algorithms reduce the execution time by an order of magnitude on power-law topologies with 1000 nodes. The reduction in memory space is similar. When there are multiple constraints, the improvement is more dramatic.