Switch scheduling via randomized edge coloring. FOCS

The essence of an Internet router is an n n switch which routes packets from input to output ports. Such a switch can be viewed as a bipartite graph with the input and output ports as the two vertex sets. Packets arriving at input port i and destined for output port j can be modeled as an edge from i to j. Current switch scheduling algorithms view the routing of packets at each time step as a selection of a bipar-tite matching. We take the view that the switch scheduling problem across a sequence of time-steps is an instance of the edge coloring problem for a bipartite multigraph. Im-plementation considerations lead us to seek edge coloring algorithms for bipartite multigraphs that are fast, decentral-ized, and online. We present a randomized algorithm which has the desired properties, and uses only a near-optimal ∆ o ∆ colors on dense bipartite graphs arising in the context of switch scheduling. This algorithm extends to non-bipartite graphs as well. It leads to a novel switch schedul-ing algorithm which, for stochastic online edge arrivals, is stable, i.e., the queue length at each input port is bounded at all times. We note that this is the first decentralized switch scheduling algorithm that is also guaranteed to be stable.