Experimental Analysis of Scheduling Algorithms for Aggregated Links

Abstract

We consider networks with aggregated links, that is networks in which physical link segments between two interconnected devices are grouped into a single logical link. Network devices supporting such aggregated links must implement a distribution algorithm responsible for choosing the transmission link for any given packet. Traditionally, in order to maintain packet ordering within conversations, all distribution algorithms transmitted packets from a given conversation on a single link. This approach is attractive for its simplicity, but in some conditions it tends to under-utilize the capacity of the link. On the other hand, due to variations in packet lengths, sending packets from the same conversation simultaneously along different physical links requires careful scheduling to ensure that the packets arrive in the correct order.

Recently, Jawor et al. [6] formulated this packet scheduling problem as an online optimization problem and proposed an algorithm called Block. The focus of the work in [6] was purely theoretical – to achieve an optimal competitive ratio. In this paper we study this problem in an experimental setting. To this end, we develop a refined version of Algorithm Block, and we show, through experimental analysis, that it indeed significantly reduces the maximum amount of time packets spent in buffers before transmission, compared to the standard methods based on hashing.