Please use this identifier to cite or link to this item:
標題: ATM網路中自相似性交通流量之有效控制-以FIR類神經網路實現
Effective Flow Control on Self-similar Traffic in ATM Networks- An FIR neural network Approach
作者: 連偉錫
Lian, WeiShi
關鍵字: ATM;非同步傳輸模式;self-similar;QoS;neural network;multiple leaky buckets;congestion control;自相似性;服務品質;類神經網路;多重漏桶;壅塞控制
出版社: 電機工程學系
非同步傳輸模式(ATM)已經成為在寬頻整合服務數位網路(B-ISDN)上傳送資料、聲音和影像的重要技術,這些媒體各有不同的服務品質(Quality of Service,QoS)要求,提供QoS保證是ATM網路必須具備的能力。 目前相關於交通流量特性的研究證實,在ATM網路中的可變位元速率之視訊,具有自相似性。 若忽略此特性,則會在網路效能分析時,產生過渡樂觀的預估值,進而導致網路資源配置不足。 換言之,自相似性對ATM網路的分析、設計及控制具有實質的影響。
類神經網路的特點在於經由學習法則的訓練之後,所選之訓練範例得以掌握資料間變化的規律。 若訓練範例具有代表性,則類神經網路將具備預測資料變化的能力。 由於自相似性訊號之片段具有與原訊號相同的統計特性且其相似的交通形式會在未來的時間內重複地出現。 因此,若將此片段當作訓練範例,則類神經網路便能對該自相似性訊號進行精確的預測。 針對上述特性,我們發展出FIR多層類神經網路來分析及預測ATM網路中自相似性交通流量的狀態。 FIR多層類神經網路能準確地預測,在下個時槽即將進入網路的總資料數。 利用此一資訊,再配合迴授式流量控制器,必能大幅降低細胞遺失率及增進網路資源使用率。
本篇論文所使用的流量控制器為多重式漏桶(MLB),它利用自相似性交通流量經由疊加之後具有統計多工增益的特性,而將同一虛擬路徑中的所有漏桶予以整合,亦即共享各漏桶的漏失率及緩衝器空間。 多重式漏桶根據各個漏桶的緩衝器佔用率,動態調整各漏桶的漏失率及緩衝器空間的大小。
為了驗證我們所提機制(MLB暨FIR多層類神經網路)的效能,我們將實際網路上的MPEG訊號及經由合成所產生的自相似性交通流量,作為系統的輸入。 模擬結果顯示,MLB暨FIR多層類神經網路所造成的細胞遺失率僅達使用傳統漏桶機制的萬分之一。
關鍵字: 非同步傳輸模式、自相似性、服務品質、類神經網路、多重

Asynchronous Transfer Mode (ATM) has been extensively applied in supporting all conceivable media including data, voice, and video in Broadband-ISDNs. For transporting such a diverse mixture of traffic sources requiring various Quality of Services (QoSs), ATM networks must offer QoS guarantees for various classes of traffic sources. Recent traffic measurement studies have demonstrated that the variable bit rate (VBR) video over ATM networks exhibits self-similarity. Neglecting this characteristic would lead to overly optimistic performance predictions and inadequate allocation of network resources. Restated, the self-similarity characteristic has practical implications for analysis, design and control of ATM networks.
A neural network can learn from experiences by providing the input-output data without the need of specifying the exact relation between the input and output. It can generalize the learned experience and obtain the correct output when new situations are encountered. All that is required are examples of the relation between the given inputs and the desired outputs. With appropriate training process, a neural network can learn such relations and produce accurate output, even when new input data are confronted. A portion of self-similar traffic data has the same characteristics as the original traffic data, and its similar traffic patterns repeat themselves in the future. Therefore, if the portion of self-similar traffic data is used to be the selected training set, the neural network can accurately predict the whole self-similar traffic. The finite-duration impulse response (FIR) multilayer network, which belongs to the time-delay neural network (TDNN), is employed to predict the number of incoming cells at the next time slot. Based on the information provided by the FIR multilayer network, our proposed feedback rate regulator can decrease the cell loss rate and significantly enhance the network resource utilization. In this thesis, the proposed feedback rate regulator for self-similar VBR traffic in ATM networks is based on the multiple leaky buckets (MLB) mechanism. The multiplexing gain is assumed herein to exist for aggregated self-similar VBR traffic. Therefore, in contrast to the conventional leaky bucket (LB), the leaky rate and buffer capacity of all LBs are shared in the same virtual path in order to more effectively use network resources. In MLB mechanisms, the leaky rate and buffer capacity of each LB are dynamically adjusted based on the buffer occupancy.
To validate the performance of our mechanisms (MLB with FIR) , ten real world MPEG1 traffic traces and synthesized self-similar data series are used in our experiments. Simulation results verify that the cell loss rate has an improvement over the conventional leaky bucket method by more than ten thousand times.
Key words : ATM, self-similar, QoS, neural network, multiple
leaky buckets, congestion control.
Appears in Collections:電機工程學系所

Show full item record
TAIR Related Article

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.