A study on traffic control in high-speed multi-service networks

我们建立了分析具有多个马尔可夫到达流的排队的算法方法，该方法可以以任何期望的精度表示任意平稳点过程。尽管这类排队在应用中是非常重要的，但它的分析一直被认为是非常困难的。在本研究中，我们证明了联合排队长度分布是以逗留时间分布的形式给出的，并在此基础上建立了算法。这一结果被认为是定量评估可容纳可变长度分组的多业务网络性能的基础。此外，我们还证明了，如果在网络边缘通过峰值速率执行器来调节长距离相关流量，那么就有可能消除由长距离相关流量引起的极端拥塞。因此，通过对服务网络进行定量建模。多马尔可夫到达流的建模方法仍在研究中。此外，通过一个简单的排队模型，我们证明了吞吐量和时延都可以通过优先级排队和动态缓冲区管理来控制。

We established the algorithmic method to analyze queues with multiple Markovian arrival streams, which can represent any stationary point process with any desired accuracy. Even though this class of queues is very important in applications, its analysis had been considered to by very hard. In this study, we showed that the joint queue length distribution is given in terms of the sojourn time distribution, and based on it, the algorithmic approach was established. This result is considered as a basis to evaluate the performance of multi-service networks quantitatively, which accommodate variable-length packets. Furthermore, we showed that it is possible to eliminate somehow extremal congestion caused by long-range dependent traffic if long-range dependent traffic is regulated at the edge of networks by a peak rate enforcer. Therefore, by modeling service networks quantitatively. The modeling method with multiple Markovian arrival streams is still being investigated. Moreover, using a simple queueing model, we showed that both the throughput and delay can be controlled by priority queueing and dynamic buffer management.