孫雅麗臺灣大學:資訊管理學研究所林雅華Lin, Ya-HuaYa-HuaLin2007-11-262018-06-292007-11-262018-06-292007http://ntur.lib.ntu.edu.tw//handle/246246/54364以802.16網路作為802.11網路後置網路(backhaul network)的交互網路裡,SS扮演著連接兩個異質網路的角色,為了減少SS和BS之間的訊息交換並有效地使用資源滿足802.11網路的傳輸需求,我們提出一套SS端的動態資源管理機制。利用Decision Period(DP)的概念令SS以多個frame為單位向BS提出頻寬需求,減少SS和BS之間的溝通頻率;並導入收入-成本模型,以收入和成本的觀念來牽制SS對資源的使用,一方面保留資源用以傳輸封包來獲取服務收入,一方面利用保留成本避免SS為增加收入而無限制地保留資源,利用延遲服務成本避免SS為減少保留成本而累積封包不傳輸的行為。由於SS是以DP為單位作決策,我們提出三種預測方法來預測未來一個DP每個frame之封包新進量,作為資源保留決策的參考。由於每次保留決策影響所及的範圍是一個DP,考慮到一個DP裡時間資源的不可儲存性和封包可無限往後累積的特性,我們的保留決策是基於三點來作考量-(1)減少服務被延遲的可能性;(2)增加已保留資源的使用率;(3)減少服務成本。 實驗結果顯示,無論來源端的類型為何,根據三種方法作預測所得的淨收入與已知實際封包新進量的最佳決策結果之淨收入都相去不遠,而且以DP為單位作決策的結果之平均淨收入幾乎總是比frame-by-frame作決策的結果還高。In the internetworking of 802.16 and 802.11 networks, the 802.16 SS plays the role connecting these two heterogeneous networks. To reduce the amount of messages exchanged between SS and BS and effectively use the resource to satisfy the requirement of packet transmission in 802.11 networks, we propose a dynamic resource management mechanism on SS. To reduce the communication frequency and message amount between SS and BS, instead of requesting resource frame by frame, we introduce the concept of Decision Period (DP)which consists multiple frames (e.g., a DP contains m concecutive frames) and SS requests resource from BS every m frames. We introduce the Income-Cost Model which uses the idea of service fee and reservation/block-off cost to evaluate the performance. In this model, the SS uses the reserved resource to serve packets in order to take in income. It will be charged with the block-off cost for each packet it fails to serve. Because the SS can not know the future arrival traffic in advance, it may conservatively reserve more resources than backlogged requirement, to the extreme, it may reserve the maximal allowable resources at each DP. However, it can not arbitrarily reserve all the resources because they are per unit charged with the reservation cost. We have to predict the new-arrival traffic of multiple frames in next DP and make reservation decision based on these predictions. Because a single reservation decision applies to multiple frames (i.e., a DP), the difficulty of our problem is that: the reserved time resource is unstorable (it will be wasted if reserved and not used) but the unserved packets will be accumulated infinitely. We propose three approaches to predict the amount of new-arrival traffic of each frame in next DP. Our reservation decision is based on three objectives-(1)minimize request blocking probability;(2)maximize resource utilization;(3)minimize cost。 According to experimental results, for all kinds of source traffic considered, the net profit from reservation decision based on the prediction of three approaches all close to that from optimal reservation decision based actual arrival. Also, in most cases, our mechanisms achieve higher average net profit per frame with decision on the basis of DP than that with frame-by-frame decision.口試委員會審訂書 i 誌謝 ii 論文摘要 iii Thesis abstract iv 目錄 vi 表次 viii 圖次 ix 第一章 序論 1 第一節 研究背景 1 第二節 IEEE 802.16 簡介 3 I. 媒體存取控制層協定與訊息 3 II. 服務類型(Scheduling Service Class) 4 第三節 研究動機 5 第四節 研究目標 10 第五節 論文架構 12 第二章 文獻探討 13 第一節 802.16網路資源管理機制 13 I. BS端資源管理方法 13 II. SS端資源管理方法 15 第二節 預測工具-Kalman Filter 17 第三節 網路流量預測(Traffic Prediction) 19 第三章 SS端動態資源管理機制 21 第一節 問題定義與假設 21 第二節 系統架構 26 I. 決策週期(Decision Period) 26 II. 收入-成本模型(Income-Cost Model) 28 第三節 方法描述 32 I. 觀測資源需求 33 II. 預測資源需求 33 III. 決定資源保留數量 42 第四章 效能分析與比較 48 第一節 實驗方法 48 第二節 網路拓樸與參數設定 48 第三節 實驗目的 50 第四節 實驗結果 53 I. 預測準確度評估 53 II. 資源保留機制整體效果評估 57 III. 對突發網路流量(burst)的預測與整體決策效果 62 IV. 參數關係 67 第五章 結論與未來展望 74 參考文獻 76 簡歷 784449162 bytesapplication/pdfen-US802.16資源管理資源保留週期收入成本流量預測resource managementresource reservationperiodincomcosttraffic prediction在802.11和802.16交互網路環境下保證服務品質的動態需求預測與資源保留機制Demand Prediction and Resource Reservation for QoS Internetworking of 802.11 and 802.16 Networksotherhttp://ntur.lib.ntu.edu.tw/bitstream/246246/54364/1/ntu-96-R94725015-1.pdf