This is the authors' abstract. We don't add key points for this paper.
We study the optimal scaling of the expected total queue size in an n×n input-queued switch, as a function of the number of ports n and the load factor ρ, which has been conjectured to be Θ(n/(1−ρ)) (cf. [15]). In a recent work [16], the validity of this conjecture has been established for the regime where 1−ρ=O(1/n2). In this paper, we make further progress in the direction of this conjecture. We provide a new class of scheduling policies under which the expected total queue size scales as O(n1.5(1−ρ)−1log(1/(1−ρ))) when 1−ρ=O(1/n). This is an improvement over the state of the art; for example, for ρ=1−1/n the best known bound was O(n3), while ours is O(n2.5logn).
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Shah et al. (2016) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: