Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/72186
Title: | 階層式雲端SDN動態資源分配最佳化 A Hierarchical Edge-Cloud SDN Controller System with Optimal Adaptive Resource Allocation for Load-Balancing |
Authors: | Frank Po-Chen Lin 林柏呈 |
Advisor: | 蔡志宏(Zsehong Tsai) |
Keyword: | 階層式SDN,雲端運算,網路模型,排隊理論,分配演算法,動態最佳化, Hierarchical SDN,cloud computing,network modeling,queuing theory,load-balancing,optimization, |
Publication Year : | 2018 |
Degree: | 碩士 |
Abstract: | 不可否認,傳統架構的網路應用既不能提供足夠的彈性,也不能真實地反映整體網路狀態。基於傳統交換機或路由器的架構不足以滿足服務品質(QoS)要求,且無法有效地處理巨大的資料量和異質的網路結構。OpenFlow協定下的軟體定義網絡(SDN)是一種新的網路設計並對網路提供一種全新的運作方式。它將網路控制平面和數據平面分離,減輕了網路控制和管理的繁重工作量。儘管SDN帶來許多的便利,然而,這樣的結構卻因為計算資源受限而在高流量網路下存在高度延遲的問題。
為了增強網路可擴展性並減少SDN網路上的計算延遲以滿足QoS要求並公平資源分配的繁重流量負載,本研究分為三個主要階段。首先,一個階層式的Edge-Cloud SDN(HECSDN)網路架構被提出。藉由本地端和雲端運算的資源整合,可以有效的減少網路設備消耗的計算資源。第二階段針對第一階段提出的網路架構設計排隊理論模型。此網路模型能使網路設計人員能夠快速估算其設計的效能,因而減少大量時間進行昂貴的實驗設置。第三,本研究藉由第二階段所提出的排隊理論模型推導出一種有效率的負載分配演算法,此演算法可以滿足HECSDN網路結構下,不同應用的QoS要求並同時達到公平性的資源分配。 本研究使用模擬的方式,評估HECSDN系統在不同網路環境下所能達到的效能,並與貪婪演算法在相同的系統架構下進行比較。 HECSDN系統在模擬結果中證明其在暫態反應時間方面的良好性能,在網路流量大幅改變時快速穩定系統,並有效縮小系統中的網路延遲時間。此外,HECSDN也在模擬中被證實具有支持大規模SDN網絡的強大性能。 Admittedly, network applications based on the conventional network architecture could neither afford enough flexibility nor truthfully reflect the overall network status. The conventional switch- or router-based architecture is equally unable to satisfy Quality of Service (QoS) requirements since it cannot efficiently deal with the huge information and heterogeneous network structures. OpenFlow-based Software Defined Network (SDN), a new network paradigm, can offer a promising approach. It decouples the network control plane and the data plane, easing the heavy workload of the network control and management[1-3]. However, the possibility that a computation-resource limited controller is congested by heavy flows still exists. In order to enhance network scalability and reduce computation delay on SDN networks for heavy traffic loads for QoS requirements and fair resource allocation, this research makes three major phases. First, a Hierarchical Edge-Cloud SDN (HECSDN) network architecture can provide great aid for cloud computing. By sharing computational effort in the cloud, the network architecture can provide a solution that could efficiently reduce the computational consumption produced by devices. The second phase is to design a queuing model of the proposed network architecture. A model description of a networking architecture enables the network designers to quickly estimate the performance of their design without spending considerable time for expensive experimental setups. Third, deduced from the queuing model, an efficient load-balancing algorithm satisfying QoS requirements and fairness allocation for different applications in the HECSDN network architecture is proposed. The QoS guarantee in this work refers to providing sufficient resource to different applications, so that the packet overdue ratio can be maintained below a certain threshold. This research uses simulation programs to evaluate the performance of the HECSDN system on the proposed edge-cloud SDN system with different parameters and compare with the state-of-art Greedy method under different patterns of traffic arrivals in the same system architecture. HECSDN has demonstrated its good performance on responsive transient reaction time for large alteration in traffic arrivals and effectively narrows down the flow delay time in the SDN system. Moreover, HECSDN performs strong capability supporting large-scale SDN networks. For instance, of the arrival flows with at least 1656 switches satisfies the QoS under fair resource allocation satisfying the Minimax criteria. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/72186 |
DOI: | 10.6342/NTU201803740 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 電信工程學研究所 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-107-1.pdf Restricted Access | 2.52 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.