A Clustering Scheme for Transparent and Scalable Bandwidth Management in High Speed Ethernet Environment

隨著網路服務品質(QoS)受到越來越多的重視，頻寬管理器在目前的網際網路逐漸變成一項重要的網路設備。而在可供使用的頻寬越來越多、實施頻寬管理時必要的封包處理越來越複雜的情況下，頻寬管理器必須提供更高的處理能力。在這篇論文中，我們提出了一個叢集式架構的方案，藉由集成多個頻寬管理器來提供較高的效能和精確的頻寬管理。通過此叢集的網路流量會被自動分散到各個頻寬管理器，而流量再分配的機制讓所有的頻寬管理器達到負載平衡。此一架構也提供了容錯機制，並容許動態增加此叢集內頻寬管理器的數目而不用停止整個系統。另外，因應叢集式的架構，我們設計了一個分散式的策略調整機制，計算所通過的流量比例來分配頻寬限制以達到正確的頻寬管理。最後從實作和實驗結果中，驗證出所提出的方案可以應用在具有透通性和延展性的高效能頻寬管理器。As quality of service gains more and more attention, bandwidth controllers become gradually one of the most important network devices in modern Internet environments. The demand for high performance bandwidth controllers is driven by the growing bandwidth available and the more complex packet processing functions that are essential in bandwidth management. In this thesis, we propose a clustering scheme by aggregating several devices to provide high throughput and implement the bandwidth control functions over a cluster. The proposed scheme makes incoming traffic self-dispatched and applies traffic redistribution to keep the load of devices balanced. Besides, the cluster can tolerance failures of devices and be expanded dynamically without shutting down the system. Based on the clustering scheme, we design a distributed policy adjustment mechanism, the proportional bandwidth allocation mechanism, to support accuracy bandwidth management. The experiment results suggest the proposed scheme can be applied to transparent and scalable bandwidth controller with high performance.1 Introduction 1 2 Related Works 5 2.1 Performance Improvement and Load Dispatching 5 2.1.1 Clustering with Traffic Dispatchers 6 2.1.2 Clustering with Self-Dispatching Mechanisms 7 2.2 Linux Netfilter/Iptables Project 8 2.2.1 Introduction 8 2.2.2 Netfilter Architecture 9 2.3 Linux Traffic Control 10 2.3.1 Classless Queueing Discipline 11 2.3.2 Classful Queueing Discipline 12 3 Clustering of Bandwidth Controllers 15 3.1 System Architecture 15 3.2 Traffic Dispatching Mechanism 17 3.3 Clustering with Traffic Redistribution 20 3.4 Fault-Tolerance and Dynamic Expansion Mechanism 24 3.5 Proportional Bandwidth Allocation Mechanism 27 3.6 Summary of the Proposed Scheme 29 4 Implementation and Performance Evaluation 33 4.1 Implementation and Experiment Environment 33 4.2 Preliminary Experiments 35 4.2.1 Accuracy of the Queueing Discipline 35 4.2.2 Throughput Drop Due to Increased Rule Count 36 4.3 Throughput of the Cluster 37 4.4 Effectiveness of the Proposed Mechanisms 38 5 Conclusions and Future Works 43 5.1 Conclusions 43 5.2 Future Works 44 References 4