On Queue-Size Scaling for Input-Queued Switches

On Queue-Size Scaling for Input-Queued Switches
复制标题

关于输入队列交换机的队列大小缩放

DOI:
--
复制
发表时间:
2014
期刊:
arXiv.org
影响因子:
--
通讯作者:
Y. Zhong
Y. Zhong
中科院分区:
--
文献类型:
--
作者:
Devavrat Shah;J. Tsitsiklis;Y. Zhong

文献摘要

被引文献

相似文献

我们研究了在一个$n中期望总队长的最优缩放 乘以n$输入排队交换机,作为端口数n$和负载因子的函数 ho$,它已被证明是$Theta(n/(1- ho))$。在最近的一项工作中,这一猜想的有效性已经建立了制度,其中1- 时间复杂度为O(1/n^2)。在本文中,我们在这个猜想的方向上取得了进一步的进展。我们提供了一类新的调度策略,在该策略下,期望的总队列大小为$O(n^{1.5}(1- ho)^{-1}log(1/(1- (ho)$当$1- 时间复杂度为O(1/n)。这是对现有技术的改进;例如,对于$ ho = 1 - 1/n$最好的已知界限是O(n^3)$,而我们的是O(n^{2.5}log n)$。
We study the optimal scaling of the expected total queue size in an $n imes n$ input-queued switch, as a function of the number of ports $n$ and the load factor $ ho$, which has been conjectured to be $Theta (n/(1- ho))$. In a recent work, the validity of this conjecture has been established for the regime where $1- ho = O(1/n^2)$. 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 $O(n^{1.5}(1- ho)^{-1}log(1/(1- ho)))$ when $1- ho = O(1/n)$. This is an improvement over the state of the art; for example, for $ ho = 1 - 1/n$ the best known bound was $O(n^3)$, while ours is $O(n^{2.5}log n)$.