請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/20476
標題: | 異質性霧運算網路之虛擬網路嵌入 Virtual Network Embedding in Heterogeneous Fog Networks |
作者: | Yu-Hsiang Chao 趙禹翔 |
指導教授: | 周俊廷(Chun-Ting Chou) |
關鍵字: | 霧運算,虛擬網路嵌入,異質性網路, Fog computing,Virtual network embedding,Heterogeneous network, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 在未來的5G甚至是次世代網路中,網路部署的靈活性和服務的多樣性是最為關鍵的兩個要求和議題。霧運算以高度分散式的形式將雲計算擴展到網路的邊緣,提供低延遲的運算服務,成為實現未來網路架構的潛在解決方案之一。作為霧運算的實現技術─網路功能虛擬化(NFV)允許多個客製化或是自定義虛擬網路(VN)共存於基礎設施提供商(InP)所擁有的同一實體網路(SN)上。在實體網路上映射虛擬網路是虛擬化環境中的一個關鍵問題,被普遍稱為虛擬網路嵌入(virtual network embedding, VNE)。
過去討論虛擬網路嵌入的問題通常將通用處理器(general purpose processor, GPP)作為能夠處理所有虛擬網路的實體運算節點,然而真實情況當中,通用處理器並不適合處理所有的運算以及網路功能,如果使用了它可能會消耗更多計算資源。例如:圖形處理、硬體解碼等。此外,在實際場景中,霧運算網路中確實存在著多種不同的特定功能節點,例如圖形處理單元(graphics processing unit, GPU),張量處理單元(tensor processing unit, TPU),硬體解碼器(hardware decoder, HD),存儲單元(storage),而非僅僅只有通用處理器的存在。現有的虛擬網路嵌入模型沒有考慮這些特定功能節點的存在與限制條件,也無法滿足實際情況的需求。 本論文提出一個基於粒子群最佳化(Particle Swarm Optimization, PSO)的增強演算法,以解決異質性霧運算網路中虛擬網路嵌入問題,此演算法稱為VNE-HF。該演算法旨在成功嵌入虛擬網路,同時減少實體資源成本的消耗,目標為在固定且相對短的時間內找到接近最佳的解決方案。模擬結果顯示,在小型網路中,該算法之長期收益優於傳統的粒子群最佳化演算法10%,在大型網路中,該算法之長期收益優於傳統的粒子群最佳化演算法23%。 In the future 5G or even beyond 5G networks, network deployment flexibility and service diversity are two of the most critical requirements and issues. Fog computing has been proposed as a potential solution to enable the desired future network environment. It extends cloud computing to the edge of the network in a highly distributed manner. As an enabler of fog computing, network function virtualization (NFV) allows multiple customized virtual networks (VNs) to co-exist on the same substrate network (SN) owned by an infrastructure provider (InP). Mapping VNs on an SN is a critical issue in a virtualized environment and is referred to as virtual network embedding (VNE). The existing VNE models consider general purpose processors (GPPs) as the substrate nodes, which are able to support all varieties of VNs. However, GPP is not suitable for all kinds of network functions, such as graphics processing, hardware decoding, which consume more computing resources. Moreover, in the real scenario, there are multiple different functional nodes, such as graphics processing unit (GPU), tensor processing unit (TPU), hardware decoder (HD), storage co-existing in the fog network, instead of only GPP. The existing VNE models do not consider these task-specific nodes’ constraints and do not keep up with the need of such a heterogeneous fog network. In this thesis, we propose an algorithm to solve the VNE problem in heterogeneous fog networks based on particle swarm optimization (PSO), called VNE-HF. This algorithm aims to embed the VNs successfully with less computing cost and bandwidth cost. The goal is to find a near-optimal solution in a fixed and rather small amount of time. The simulation results show that the proposed algorithm outperforms the traditional PSO algorithm in terms of long-term average revenue by 10% and 23% in small and large heterogeneous fog networks, respectively. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/20476 |
DOI: | 10.6342/NTU202004177 |
全文授權: | 未授權 |
顯示於系所單位: | 電信工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-2708202001542500.pdf 目前未授權公開取用 | 3.12 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。