臺灣博碩士論文加值系統

　　非同步傳輸模式網路的流量管理方法很多，但大多屬於使用者端的交通參數的控制設定，如：尖峰率、封包延遲變化、可支撐封包速率、最大突發性大小等參數影響；未有對網路流量狀況的輕重加以考慮，造成頻寬的使用率未能充份利用，亦既不能預測本身頻寬的需求，不能提供明確的服務保証。其次，在提供各種服務品質資料傳輸速率之一的“可使用位元速率”，在回饋控制方面，如預支式及速率式的流量控制，須透過回饋資訊，來指出速率應增加或減少，所以在高速網路擁塞避免、恢復等程序上，因有傳輸延遲的影響，未能即時通知發送端降低傳輸速率，以致造成網路擁塞狀況持續惡化，違反細胞數目增加，以及負載加重，導致細胞的流失。所以為了同時解決上述的問題，本研究提出了兩種預測方法，於回饋資訊到達使用者端的傳輸延遲之前，開始預測細胞速率應增加或減少；亦即利用“可使用位元速率”的回饋控制，來預測網路狀況的變化，作為動態調整隙漏水桶法的權杖產生速率。
　　首先，運用類神經網路的倒傳遞網路模式，針對網路的佇列延遲與負載量之間的關係，配合優先權控制的方法，並透過類神經網路的預測技術，以掌握佇列延遲與負載量的關係行為模式。其次利用卷積碼編碼及維特比解碼的方法，因其編碼、解碼的過程與前後時刻收到碼組的提取有關，所以根據網路擁塞狀況通知的訊息位元之變化，透過卷積碼編碼及維特比解碼的方法，來預測網路佇延遲的增減變化情形。利用上述二種方法，分別作為隙漏水桶法預測網路擁塞狀況的機制，以動態調節細胞傳輸的速率，用來控制流量的增加或減少，達到最佳的頻寬使用率及擁塞控制。透過實驗模擬結果顯示，我們所提出的兩種方法，皆能有效預測網路擁塞狀況作為流量控制的好方法。

　　Flow control methods for ATM networks are many. Most of them assume that the user side can determine the peak rate, cell delay variation, sustainable cell rate, maximum burst size, etc. One the traffic contract is determined, the user traffic enters the network accordingly. As a result, the network congestion state is not possible.
　　Secondary, The "available bit rate" is one of the service of the ATM networks. The feedback control includes credit-based and rate-based. We must point out that the rate of the user traffic should increase or decrease through the feedback information. Because the propagation delay impact on the high speed network congestion avoidance, recovery procedures, it can not notify the sender to reduce the rate and makes the network congestion condition get worse and the violation cells increase, the load becomes heavy and results in the cell lose. For solving the above problems simultaneous, we propose two predictive methods to predict the cell increase or decrease. It uses the feedback control of "available bit rate" to predict the network situation as the basis of token generation rate increase or decrease of leaky bucket.
　　First, we use the back-propagation network model of the neural network and consider the relationship between queuing delay and offered load, and utilizes priority control methods and neural network model to predict network traffic conditions, and thus maximize the network utilization. Secondary, we use the convolution encoding and Veterbi-Trellis decoding model. So according to the various states of network congestion situation, we use the convolutional coding method and Veterbi-Trellis decoding method to predict the increase/decrease of network queuing delay, and thus adjust the user data rates.