高速多媒體通訊網路之可調整視窗平均封包速率使用參數控制

Abstract

在這篇論文中，首先我們介紹乏晰訊務塑型器(TS)-使用參數控制器(UPC) ，來監控一個連線的平均封包速率。它是由兩個乏晰漏水桶法所構成的TS和UPC連結而成。此乏晰漏水桶法含有一個乏晰水增額(increment)控制器（FIC），它利用此連線的長期平均速率和短期平均速率，以及一組語言性法則，來調整水增額的量。接著我們提出增強型乏晰TS-UPC來改善原來的機制。這個增強型乏晰TS-UPC 包含了 FIC 和一個乏晰視窗估計控制器（FWE）。這個乏晰視窗估計控制器也是利用資料連線的長期平均速率和短期平均速率（在可變動視窗所量測）來決定下一個短期平均速率是該在多大的視窗中量測。藉由FWE 我們可以適當得調整短期的視窗長度，使的所量測到的短期平均速率較固定視窗大小的更為準確，及真實的反映出資料連線的情況。當然有了更適當的短期平均速率，FIC 所決定出的水增額將更為符合所需要的。水增額的調整是基於一個目標--達到較好的表現，在選擇性、反應性、和平均佇列延遲方面。模擬結果顯示增強型乏晰TS-UPC在各方面皆優於乏晰TS-UPC 尤其在平均佇列延遲方面。當然乏晰TS-UPC在各方面的效能明顯優於傳統TS-UPC。 接下來我們提出增強型的類神經乏晰TS-UPC，它包含一個類神經乏晰水增額控制器（NFIC）和上述所提的乏晰視窗估計控制器（FWE）。同樣的這個NFIC 也是採用資料連線的長期平均速率和短期平均速率來決定水增額的量，而且它有一個額外的強制訊號（reinforcement signal）使的它有線上學習（on-line learning）的能力，所以在選擇性，反應性方面會有很大的改善。而FWE 的加入使的短期平均速率的量測更為精準，對於平均佇列延遲方面的改善有很大的效果。實驗結果顯示增強型的類神經乏晰TS-UPC在選擇性、反應性、和平均佇列延遲方面比起其他的TS-UPC都呈現最好的表現。

In this thesis, we first introduce the fuzzy TS-UPC model, which is conjunctions of the TS (traffic shaper) and the UPC (usage parameter controller) for sustainable-cell-rate usage parameter control in high-speed multimediacommunications. The fuzzy TS-UPC is composed of the fuzzy leaky bucket. The fuzzy increment controller (FIC), which is incorporated with the conventional leaky bucket algorithm in fuzzy leaky algorithm, properly choose the long-term mean rate and the short-term mean rate as input variables and intelligently compute to determine the increment value. Then we propose a enhanced fuzzy TS-UPC to improve the fuzzy TS-UPC. The enhanced fuzzy TS-UPC contains the FIC and a fuzzy window estimator (FWE). The FWE chooses the long-term mean rate and the short-term mean rate (in the updated short-term window) as input variables to decide the increment or decrease degree of the next short-term window. It helps to suitably adjust the short-term window for short-term mean rate calculations and then makthe short-term mean rate more correctly reflect the traffic conditions than the fixed short-term time window. Three performance measures such as selectivity, responsiveness, and queueing delay are considered. Simulation results show that the enhanced fuzzy TS-UPC can have better performance than the fuzzy TS-UPC. And the fuzzy TS-UPC outperforms the conventional TS-UPC. Next we propose the enhanced neural fuzzy TS-UPC, which is composed of the neural fuzzy increment control (NFIC) and the fuzzy window estimator (FWE) to make the TS-UPC more robust. The NFIC chooses the long-term mean rate and the short-term mean rate as input variables and then decide a more appropriate increment value. In the NFIC, the reinforcement learning can learn on line from a reinforcement signal (the difference between desired loss ratio and measured loss ratio), so the increment value can reflect the traffic conditions. The FWE described above suitably adjusts the short-term window and then make the short-term mean rate more correctly. Simulation results show that the enhanced neural fuzzy TS-UPC achieves the best performance in selectivity, responsiveness, and queueing delay, than all other TS-UPCs.