下世代分頻多工存取被動光網路之平行化多目標基因演算法動態頻寬分配設計

Abstract

在此論文我們提出一個最佳化、快速的動態頻寬分配設計(Dynamic Bandwidth Allocation, DBA)，用於下世代分頻多工存取被動光網路(Next-Generation OFDMA-PON)上，此架構具有高頻寬、低成本、節能的優點，同時可整合無線網路訊號的傳輸，我們以多目標基因演算法達到最佳化，並使用平行化進行快速運算，能即時監控網路流量並調整頻寬分配，讓在該架構上的使用者獲得高throughput且公平性(fairness)的服務品質保證(QoS)。我們使用PQS-PR形式的媒介存取控制(Medium Access Control, MAC)，其類似但不同於傳統的token-bucket，同時將複雜的頻寬分配機制轉換為較易處理與理解之最佳化問題，利用非支配型排序基因演算法-II (Non-dominated Sorting Genetic Algorithm-II, NSGA-II)，仿效大自然適者生存的法則以取得最佳解，而訊息傳遞介面(Message Passing Interface, MPI)是現今常被採用的平行運算方案，我們用來加速處理最佳化問題的運算速度。我們期望完美結合以上幾種方法，提供在此新穎的下世代光網路架構的使用者，一套完整且最佳的動態頻寬分配與服務品質方案。

In this thesis, we propose a optimal, fast dynamic bandwidth allocation (DBA) for Next-Generation OFDMA-PON, which has high bandwidth, low-cost, energy- saving advantages and compatible with wireless signals. Our purpose is to monitor network traffic on-line and adjust bandwidth allocation to achieve high throughput and fairness to ensure quality of service (QoS). Multi-objective genetic algorithm (MOGA) is used to achieve optimization and transform complex bandwidth allocation mechanism into a more easily understanding and handled problem. Message Passing Interface (MPI) is now frequently used and we use it to expedite the processing of parallel computing speed. We expect a perfect combination of several methods to provide users the next generation optical network architecture with optimal bandwidth allocation and quality of service.