全光網際網路傳送交換及訊務之研究

Abstract

現今，分波多工（Wavelength Division Multiplexing, WDM） 設備已經十分成熟。分波多工設備已經被廣泛地用來建設高容量的光網路，俾使能順利地傳送日益增加的網際網路訊務量。因此，分波多工光網路與網際網路通訊協定（Internet Protocol, IP）直接的整合便成為最重要的研究之一。其中，全光網際網路是IP與分波多工光網路整合最終要達到的目標。本論文探討了一部份有關全光網際網路傳送、交換及訊務之研究。在第二章中，我們首先提出一個新的分波多工光網路傳送IP訊務之通訊協定。該協定可以同時處理實體層與數據鏈路層的碼框同步與封包識別，並且可以扮演一部分第三層（網路層）交換的功能。我們同時亦在本章中詳細地分析該協定的碼框包裝的成效。在第三章中，我們先介紹兩種分波多工光路由器，並且比較他們在非同步、可變封包長度的自我相似性網際網路訊務（Self-Similar IP Traffic）下的封包遺失機率成效。我們發現當迴授埠數目大於四時，使用迴授型（Feedback Type）但未利用空隙填塞（Void Filling）排程演譯法的分波多工光路由器將較使用空隙填塞排程演譯法的前饋型(Feed-Forward Type)分波多工光路由器的封包遺失率成效佳。之後，我們在該章中詳細地分析迴授型分波多工光路由器的系統大小與迴授型緩衝器特性之間的關連與交互影響，俾使能達到最佳化的設計。在第四章中，我們延伸以變異常數為基礎（Variance-Based）的馬可夫式參數找尋法（Markovian Fitting），並且使用具有短程關連性（Short Range Dependence, SRD）的馬可夫模型作為模擬自我相似性網路的訊務。在該章中，我們詳細地介紹了如何找尋所需要的馬可夫模型參數，並且利用廣泛的數值結果證明該模型的二階統計特性與排隊模型成效之準確性與時間期間對其之影響。最後，我們將在第五章總結本篇論文，並且說明未來可延伸的研究方向與工作。

Nowadays, equipments and components of wavelength division multiplexing (WDM) have matured enough to provide extremely high-capacity networks that are required to transport the ever-increasing amount of IP traffic. Dense WDM (DWDM) system has been conceived to be able to provide the extraordinarily additional bandwidth in fiber optic networks. At the present time, DWDM technologies doubtlessly are the means that can fully exploit the tremendous bandwidth of optical fibers. Therefore, the direct integration of IP and WDM/DWDM has been the most popular research and development area for the next generation Internet. Among the research of IP over WDM networks, optical router based optical Internet is the ultimate goal of integrating IP with WDM directly. In this thesis, we investigate three basic issues of optical Internet. These issues are presented in the three major chapters in this thesis. We briefly describe them as follows. In Chapter 2, we propose a framework for IP over WDM networks using a newly proposed IP over WDM protocol with routing and transport capabilities. This newly proposed protocol uses label and length information as the physical transport framing and data link delineation at the same time, and it can handle routing processes in the core networks. The proposed protocol performs physical framing, data link delineation, and part of layer 3 routing functions in one process. Thus, it can speed up the processes in future tera-bit networks. We present the analysis of framing encapsulation performance of this protocol in detail in Chapter 2. In Chapter 3, we introduce two different types of WDM optical routers: one is feed-forward (FF) type and the other is feedback (FB) type or the so-called partially shared buffering (PSB) type WDM optical routers. When WDM optical routers operate under asynchronous and variable packet length mode, there will be voids induced in the fiber delay lines (FDLs). These voids make the channels idle and are unusable for storing newly arriving packets. Thus, the performance in terms of probability of packet loss (PPL) is inferior to the routers operating under synchronous and fixed packet length mode. In this chapter, we first compare the packet loss performance of these two different types of WDM optical routers incorporating various contention resolution schemes and operating under asynchronous and variable packet length self-similar traffic input. we demonstrate that a 16X16 FB type WDM optical router employing more than 4 re-circulated ports without using void filling (VF) algorithm can provide better performance than that of FF type WDM optical routers using VF algorithm under asynchronous and variable packet length self-similar traffic input. We then investigate the system dimensioning issues of FB type WDM optical routers under the same traffic by simulation in Chapter 3. In Chapter 4, we investigate the possibility of using short range dependent (SRD) Markovian models as appropriate traffic models to emulate the second-order self-similar traffic. Research results have shown that self-similar or long-range dependent (LRD) traffic has severe impact on switch performance and network design. New traffic models such as chaotic maps, fractional Brownian motion (FBM) and fractional autoregressive integrated moving average (FARIMA), etc., have been proposed to characterize the self-similar and LRD behavior. Although these traffic models characterize the self-similar process in a parsimonious way, most of them are asymptotic in nature, hence they are less effective in the context of queueing-based performance evaluation when the buffer sizes are small. In this chapter, we proposed a generalized variance-based Markovian fitting for self-similar traffic modelling. It is found from extensive numerical results that not only the proposed Markovian fitting retains the advantages of variance-based fitting, but also its accuracy is better than that of the original work. We present the performance comparison based on the second-order statistics of counts and the queueing-based performance measures such as tail probability and loss probability in this chapter. The analysis of the time scale effect on the proposed model is also given in detail in this chapter. At last, we conclude this thesis in Chapter 5, by summarizing the works that we have achieved. There are still several important issues regarding optical Internet that we have not been available to investigate. These issues will be presented in the future works in Chapter 5.