使用Fuzzy Q-learning達成具公平性動態頻寬分配的MPLS-TP環

Abstract

近年來隨著行動上網的普及，資料形態的訊務需求日漸成長。由於消費大眾期待行動服務的收費價格能夠更低廉，電信營運商勢必在擴展網路的同時，必須兼顧降低成本支出。因此乙太網路技術夾帶著低價高效率的優勢，應用範圍逐漸從近端區域網路擴展到都會型區域網路。在使用乙太網路服務做傳輸媒介之上，提供各式功能的封包傳輸型網路之中，MPLS-TP可以支援現有各式標準的服務保證機制，且具有虛擬鏈路及保護的功能。本篇論文將以提供公平性的篇刊分配在環狀網路中為基礎，考量了現實傳輸網路中，可用頻寬為隨時間動態變化的情況。傳統的fair rate機制所面對的震盪現象，將導入fuzzy Q-learning的機制進行改善，並以差異速率作為回授訊號之用。在模擬結果中可顯示，本文提出的fair rate產生器，將可有效改善MPLT-TP環的效能，包含降低收斂時間及提高整體網路頻寬使用效率。

The data-type traffic from mobile equipment grows rapidly day by day. While consumers want to pay less for mobile services, network operators must reduce their cost when scaling networks. Since the cost and media-usage efficiency advantages, Ethernet is extending from local area network to metropolitan area network. Beyond the attachment point as the transmitting medium, Ethernet services may operate over different kinds of packet transport networks (PTNs). MPLS-TP is one of the PTN standard. The MPLS-TP adopts all of the supporting QoS mechanisms already defined within the standards, and also brings the benefits of path-based, in-band OAM and protection mechanisms found in traditional transport technologies. This thesis is base on providing the fairness property of ring network as in previous related works. However, we make the total available bandwidth for fairness eligible traffic become dynamic in MPLT-TP ring, which is real in transport network. The fair rate scheme would be face with oscillation. So we adopt the fuzzy Q-learning algorithm in fair rate generator to enhance the performance of network. We find a signal which called rate difference take it as the reinforcement signal for fuzzy Q-learning. The simulation results reveal the purposed fuzzy Q-learning fair rate generator can make the MPLS-TP ring network with lower convergence time and higher utilization.