光叢集交換網狀網路下利用跨層通訊技術之路由與波長分配機制

Abstract

光叢集交換(optical burst switching, OBS)技術具有低成本、高頻寬使用率以及較低的延遲時間之優勢，被視為下一代網際網路的解決方案。OBS中，如何適當的分配連結資源以減輕碰撞問題及避免選擇到超載連結為一相當重要的議題。本篇論文針對光叢集交換網狀網路(OBS mesh network)架構下，使用跨層通訊技術收集網路各層的資訊，設計出一種新的動態路由及頻寬分配機制，使其可以根據實際網路狀況與訊務通訊品質的要求，決定不同訊務的路由方向及所需頻寬。我們所提出的網路跨層架構包含數個元件：實體機構、抽象機構、跨層資訊提供者、決策資料庫、決策引擎。OBS網狀網路中，邊緣節點與核心節點上的決策引擎都會執行跨層路由及波長分配(cross-layer routing and wavelength assignment, CROWN)機制，此機制利用儲存在決策資料庫的跨層資訊，包括網際網路層、資料連接層與實體層的相關資訊，以及各訊務的特性，來決定要分配給每個叢集(data burst)的路徑和波長。不同於傳統的跨層通訊架構，我們所提出的網路跨層架構中訊息可在節點之間作交換，以優化網路資源的配置。此外相較於傳統架構的逐層決策，相關的決策都會在CROWN機制中執行，有助於提升決策速度。模擬結果顯示CROWN機制能做更有效率的資源分配，降低系統的叢集損失機率以及終端對終端的延遲時間，能有效提升OBS網路的系統效能。

Optical burst switching (OBS) is a promising switching paradigm that was proposed as a candidate transport solution for next generation optical Internet. Due to the bufferless transmission in OBS, the problem of burst resource contention at core nodes is inescapable, and the unresolved contention leads to burst losses. To prevent from selecting overloaded links and guarantee certain level of QoS, development of efficient routing and wavelength assignment (RWA) algorithms is important for OBS networks. In this thesis, a cross-layer routing and wavelength assignment (CROWN) scheme for OBS mesh networks is proposed. By using the cross-layer communication system, the CROWN scheme has the information of each layers and can determine appropriate route and bandwidth for data bursts. The proposed cross-layer architecture contents several components: physical entities (PE), abstraction entities (AE), cross-layer information provider (CIP), decision database (DDB), and decision engine (DE). In the networks, DE on edge node and core node is performed by the CROWN scheme. The CROWN scheme utilizes the cross-layer information stored in DDB, and the features of different service type traffics to determine the route and wavelength for each data burst. The simulation results show that the CROWN scheme can make better utilization of network resource, decrease the burst loss probability, and increase the system throughput.