具延遲敏感性的資訊串流之特性化與允入控制

Abstract

為了保證網路的服務品質，適當的頻寬配置是在設計高速網路（例如寬頻整體服務數位網路）時非常重要的一環。為了避免因配置過多的頻寬而降低網路的利用率，因此，輸入網路的資訊串流之特性賦予必須儘可能的準確。其中資訊串流的叢聚性是影響網路品質的一項重要特性。目前，此項特性並沒有一致可接受的明確定義。本論文將利用多工器的遺失機率來定義資訊串流的叢聚性。 吾人將利用所定義的叢聚性，提出一個最佳的資訊串流量化方法。此方法可以得到一個虛擬的量化資訊串流模型；由此虛擬的資訊串流模型所評估的遺失機率將是真實遺失機率的上界。而數值分析的結果也發現，經由此量化方法所導致的網路頻寬利用率的下降程度很小。 以此虛擬的量化資訊串流模型為基礎，吾人可以進一步設計一個可即時執行的允入控制方法。同時，也可一併研究個別輸入串流的遺失機率。根據個別輸入串流的遺失機率，將可導引出一個可即時執行而且保證個別輸入源的服務品質的允入控制方法。 最後，吾人可將前述的結果推廣至具有多重服務品質的高速網路。而由其理論及數值的分析也一樣可發現，藉由虛擬的量化資訊串流模型所設計的可即時執行的允入控制方法，可以正確地估計個別輸入串流的遺失機率。

Resource allocation is necessary for a network which guarantees quality of service. To avoid over allocating resources to increase network utilization, traffic sources must be characterized as accurate as possible. An important traffic characteristics is the burstiness of the traffic stream, which, unfortunately, does not have a unanimously acceptable definition. In this thesis, we provide a definition of burstiness based on the loss probability at a bufferless multiplexer. The definition is aimed at characterizing delay sensitive traffic streams which do not allow delay caused by buffering. This definition is utilized to characterize (or quantize) traffic streams for networks which vrovide only finitely many bit rates for users to describe their traffic. To guarantee quality of service,the quantized version has to be burstier than the real traffic. We present an optimum quantization algorithm which yields a quantized version that is burstier than the real source and achieves minimum loss rate for all possible allocated bandwidths. The quantized version, called a pseudo source, can be used to represent the real source and is considered by the network for bandwidth allocation. Using the optimum quantization algorithm,we design a real-time admission control scheme for ATM networds. Both global loss probability and individual loss probability are considered in the proposed admission control schemes. To cope with multiple quality of service requirements,the concept of burstiness is generalized to the case which the cells of each traffic stream are allowed to have different loss priorities. We prove the optimum quantization algorithm we proposed for single quality of service requirement environments is also applicable to this case. By using this property, we design an efficient admission control scheme for heterogemneous delay sensitive traffic with multiple quality of service requirements. Numerical results that our proposed admission control schemes are efficient.