高速網路中使用乏晰╱類神經技術與預測變數之訊務控制

Abstract

近年來，網際網路的人口急速增加。網際網路上的應用從數據資料到語音及影像，各種應用所需要之服務品質要求也不同。非同步傳輸模式(ATM)具有整合各種通訊服務的能力，故被採用為高速網路整合各種應用服務的關鍵技術。因為訊務的多樣性及所要求的服務品質不同，網路服務提供者需要一個精巧且即時的訊務控制器來確保所有線上用戶的服務品質，並且希望能進一步提高整體網路的使用效益。 智慧型控制機制在這幾年來被廣泛應用在許多訊務控制相關的問題上。根據研究顯示，智慧型控制機制比傳統的方法有更好的表現。因此我們採用智慧型控制技術去實現高速網路上所需的訊務控制器。 在此篇論文中，我們提出了使用預測變數的乏晰╱類神經乏晰的允諾控制器。為了能反應新連線對系統的影響，我們除了使用新連線所宣告的訊務特性參數與服務品質參數外，也運用平行迴路類神經網路(pipelined recurrent neural network)預測系統擁塞情況和封包遺失率來決定是否接受新使用者建立新連線要求，使該允諾控制器即使在網路訊務變動劇烈時，也能依預測結果，事先做出適當的處理，使線上用戶都能獲得服務品質的保證並提高整體網路的使用效益。

In recent years, the number of Internet people grows rapidly. Internet applications are ranging from non-real time data services to real-time audio and video service, which need different quality of service (QoS) requirements. Asynchronous transfer mode (ATM), integrated services (IntServ), and differentiated services (DiffServ) have been proposed as the techniques to integrate diverse application services for high speed networks. Because of diverse traffic sources and associated QoSs, a network requires a sophisticated and real-time traffic controller to guarantee QoSs for existing calls and to achieve high system utilization. Recently, intelligent techniques (e.g. neural network, fuzzy logic) have been widely applied to deal with traffic control related problems in networks. Research results showed that intelligent techniques had better performance than conventional schemes. In this thesis, we propose a predictive fuzzy/neural-fuzzy connection admission controller. It employs the offered traffic characteristics, QoS requirements, and predicted network operation performance measures to decide whether to accept a new call setup request or not in order to capture the effects well after the new call is established. A pipelined recurrent neural network is used as a predictor to attain the advance information of the system.