Research on the QoS Control of Next Generation IP Network

(1)TCP/IP路由器的服务质量控制研究本研究的目的之一是提出一种新的路由器分组控制算法，解决V-WFQ问题。在V-WFQ中，如果TCP和UDP连接在路由器上聚合，吞吐量的公平性会降低，如果一些连接在路由器上加入或离开，则队列长度会变长。为了解决这些问题，我们提出了一种使用连续计数器和删除计数器来统计当前活跃连接数的算法。保证丢包概率的快速连接接纳控制研究我们研究了稳态下队列长度的尾概率的指数衰减性，并给出了指数衰减性的充分条件。得到了一个定理：如果队长分布函数的Laplace-Stieltjes变换收敛轴的实点是极点，则尾概率呈指数衰减。这个结果是如此普遍，以至于它可以应用于许多类型的队列。(3)分组网络中延迟时间的评估我们研究了分组端到端延迟时间的确定性上界。假设业务流由令牌桶算法整形。在这项研究中，我们提出了一个新的上界，它比其他传统的上界更精确，因为我们注意到聚合的突发是随着时间的推移而被处理的。通过计算机模拟，我们证实了我们的上界优于其他上界。

(1)Study on the QoS control at TCP/IP routerOne of the purposes of this research is to propose a new algorithm of packet control at a router which solves problems of V-WFQ. In V-WFQ, the fairness of throughputs degrades if TCP and UDP connections are aggregated at a router, and the queue length becomes long if some connections join or leave at a router. To solve these problems, we propose an algorithm which uses continuous counter and deleting counter to count the present number of active connections. We confirmed the performance of our method compared to other conventional methods.(2)Study on a fast connection admission control which guarantees the packet loss probabilityWe investigate the exponential decay of the tail probability of the queue length in the steady state and give a sufficient condition for the exponential decay. A result obtained in this research is a theorem that if the real point of the axis of convergence of the Laplace-Stieltjes transform of the distribution function of the queue length is a pole, then the tail probability decays exponentially. This result is so general that it can be applied to many type of queues. The result of this research was submitted to the IEEE transactions.(3)Evaluation of delay time in a packet networkWe study a deterministic upper bound for the end-to-end delay time of a packet. It is assumed that a traffic stream is shaped by a token bucket algorithm. In this study, we proposed a new upper bound which is more accurate than other conventional bounds by noticing that the aggregated burst is processed as time goes. We confirmed that our upper bound is superior to other bounds by computer simulation.