Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/51612
標題: 周期性重組光封包交換之模型化及效能分析
Modeling and Performance Analysis of Optical Packet Switching with Periodic Reconfiguration
作者: 林偉 
關鍵字: 基礎研究;資訊工程--硬體工程
摘要: 
近來光交換技術成為另一種取代電子交換器之方式。光交換器與電子交換器之運作方式相當不同。光交換器需時時地經由機械式設定及同步來達成目的,而其延遲相當地顯著,通常為幾十個奈秒至幾個毫秒。在進行重組時,光交換會暫時停止封包交換將它們存於緩衝器。重組增加了除輸出競爭所造成之額外延遲。在此計畫,我們指出重組頻率於重組延遲及排隊延遲有關鍵性的影響力。降低重組頻率有助於重組延遲之減少,然而卻增加封包於佇列的停留時間及浪費交換器於輕載時的頻寬。另一方面,提高重組頻率雖可降低封包於佇列的停留時間,卻增加了整體重組延遲時間。因此,我們提出一個針對重組頻率對光交換器效能影響的分析研究。我們想要用分析結果去找到光交換器在不同工作情況下之適當重組頻率,進而調適出交換器之最佳效能。之前,已有一些方法被提出用來減少重組延遲。但這些方法利用決定式排程方式將封包映射至最小集合之組態,只能達成一群之局部最佳化。在此我們提出另一種不同處理方法去研究這類效能最佳化問題。我們將交換器重組當成是一種包函若干個特徵化封包進入排隊隨機變數之隨機處理。而我們所追求的是統計上的最佳化。由於重組延遲及虛擬輸出佇列之加入,常見之佇列模式完全地無法適用於我們所考慮的光交換器效能分析。我們提出發展一個新的具有限時規範及變動停滯時間的離散式排隊系統。我們將利用隨機分解去建置巢狀馬可夫鏈,用之分析交換器效能。我們打算經由分析求出完整式去量化主要系統參數對於交換器效能之影響。此計畫書描述一個用來從完整式產生數值結果之計算方法。我們利用數值結果將可找出最佳的重組週期時間,用最小的延遲時間去傳遞封包。我們將會製作一個用以驗證分析結果正確性之模擬器。我們所提出的研究計畫對於光封包交換器之效能提升有相當重大的意義。在不同的負載狀態下,我們可利用所找到最佳的重組頻率,動態地優化光交換器之效能。

Recently, optical switching technology is becoming an attractive alternative to electronicswitches. Optical switches operate in a very different way than their electronic counterpart.They are periodically reconfigured through mechanical settling and synchronization, with asignificant delay ranging from tens of nanoseconds to several milliseconds. Duringreconfiguration, an optical switch temporarily suspends packets transmission and holds themin buffers. Reconfiguration causes extra delay for packets departure in addition to the queueingdelay due to output contention. In this proposal, we address the issue that reconfigurationfrequency in fact has a crucial effect on reconfiguration delay and queueing delay. Loweringreconfiguration frequency helps reduce the reconfiguration delay on average, but this mayprolong the queueing time of packets in buffers and waste switch bandwidth especially underlight load. On the other hand, raising reconfiguration frequency cuts down the queueing timeof packets, but it increases the reconfiguration delay in all. Therefore, we propose toanalytically study the impact of reconfiguration frequency on the optical switch performance.We intend to use the analytical results to find the proper frequency for optical switches tooperate with peak performance.In the past, several reconfiguration schemes have been proposed in an attempt tominimize reconfiguration delay. The main idea behind these schemes is to group arrivingpackets in one batch at a time and to deterministically schedule the packets with a minimumset of configurations. These deterministic scheduling schemes, however, can only achieve alocal optimization for one single batch. Here we propose a different approach to investigatingthe performance optimization problem. We consider switch reconfiguration as a stochasticprocess with random variables that characterize packets arrival and queueing. In other words,we are looking for statistic optimization instead. With the introduction of reconfiguration delayas well as VOQs, the conventional queueing models, e.g. Geox/G/1, flatly unfit for theperformance analysis of optical packet switches under consideration. Here we propose todevelop a new discrete-time queueing system that observes the time-limited service disciplinewith variable vacation time. By means of stochastic decomposition, we will construct ananalytical model with nested Markov chains and proceed to analyze the switch performance.Through the analysis, we will be able to quantify the impact of major system parameters onthe switch performance in closed forms. In this proposal, we sketch a computational methodfor generating numerical results from these closed forms. Using the numerical results, we canspecifically locate the optimum reconfiguration frequency at which optical switches transmitpackets with the minimum delay time. We will implement a simulator for validating theanalytical results produced by the queueing system. The proposed research is of greatsignificance to the performance enhancement of optical packet switches. With the finding ofoptimum frequency, we are capable of dynamically tuning up the performance of opticalswitches in response to varying loading conditions.
URI: http://hdl.handle.net/11455/51612
其他識別: NSC100-2221-E005-066
Appears in Collections:資訊科學與工程學系所

Show full item record
 
TAIR Related Article

Google ScholarTM

Check


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