應用於OPMDC光資料中心網路之可動態調適資料流排程技術

Abstract

為因應多樣化且資料量龐大的應用服務，我們之前已設計並建構了新一代光纖資料中心架構－OPMDC（Optical Pyramid Data Center）。不同於傳統電資料中心，其具有高擴充性、高吞吐量、低延遲、低耗能等特性。另外，我們並針對靜態的資料流環境設計了光路排程，獲致很好的成效。然而，在真實的環境中，靜態排程並無法處理快速變動的資料流型態。此外，WSS光交換機交換延遲動輒數十毫秒，頻繁的交換動作將嚴重增加整體的傳輸延遲。 本論文的研究目標即為針對OPMDC架構，延伸設計一套可根據資料中心即時網路資料流狀態而動態調適的光路排程及分配機制，以達到高吞吐量及低交換延遲的目標。此排程機制會根據目前所偵測(monitor)到的局部及整體的流量狀況，並考慮最少交換次數等條件，自動做出最佳的光路調整。模擬實驗結果顯示，在不同的網路流量狀態下，此光路排程分配機制可讓OPMDC達到低延遲且相對好的吞吐量表現。

For growing various application services, we have designed and implemented the new generation optical data center architecture- OPMDC (Optical Pyramid Data Center). Different from traditional electrical data center, OPMDC is characterize by high scalability, high bandwidth, low latency and low power consumption. In addition, we designed the wavelength scheduling for static traffic, which makes OPMDC perform well. However, the static-traffic scheduling cannot cope with dynamic traffic pattern in the practical environment. Also, the WSS-based optical switch takes tens of micro-second to switch wavelength assignment, so frequent wavelength allocation update will severely damage OPMDC throughput performance. Our study is based on the OPMDC architecture, to design a wavelength scheduling, which can automatically adjust wavelength allocation to different traffic patterns for reaching good throughput performance, considering switching time caused from optical switch. From the simulation result, under different traffic patterns, the OPMDC works by the wavelength scheduling we design can reach very low delay and good throughput.