設計與實作基於OVS-DPDK並結合硬體加速的適應性流量卸載策略;Design and Implementation of an Adaptive Flow Offloading Strategy with Hardware-Accelerated OVS-DPDK

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/95595

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95595

Title:	設計與實作基於OVS-DPDK並結合硬體加速的適應性流量卸載策略;Design and Implementation of an Adaptive Flow Offloading Strategy with Hardware-Accelerated OVS-DPDK
Authors:	李兆偉;Lee, Chao-Wei
Contributors:	資訊管理學系
Keywords:	軟體定義網路;封包處理;硬體卸載;智慧型網路卡;Open vSwitch;SDN;Packet Processing;Hardware Offloading;SmartNIC;Open vSwitch
Date:	2024-07-29
Issue Date:	2024-10-09 17:05:18 (UTC+8)
Publisher:	國立中央大學
Abstract:	現今雲端資料中心結合軟體定義網路 (Software-Defined Networking, SDN)、網路功能虛擬化 (Network Function Virtualization, NFV) 與硬體卸載 (Hardware Offloading) 等技術，能夠以可程式化網路方式來強化資料中心的網路架構和加速封包處理過程，使網路基礎設施能承載更多的流量、以及具備更穩定的運作狀態。Open vSwitch (OVS) 是一種SDN/NFV架構下常見的虛擬交換器，能利用智慧型網路卡 (SmartNIC) 將轉發規則卸載至硬體轉發表，進一步加速封包轉發的效率。然而硬體轉發表空間有限，需建立合適的卸載策略以充分利用硬體轉發表的空間。然而一般研究主要是基於固定轉發規則卸載數量上限的同時，盡可能地將所有合適的轉發規則進行卸載與替換，因此無法根據當前網路流量的變化，動態調整所需使用的硬體轉發表空間。本研究將以資料中心網路的運作情境為背景，結合網路虛擬化架構的軟硬體元件，提出一流量感知的適應性流量卸載 (TA-AFO) 策略。TA-AFO透過持續監控OVS-DPDK上的流量變化，動態調整觸發轉發規則卸載到SmartNIC的門檻值，以減少硬體空間的消耗並確保轉發效率。除此之外，亦將TA-AFO整合至OVS-DPDK中，並使用Nvidia BlueField2智慧型200G網路卡進行驗證。實驗結果顯示，在搭配真實網路流量的模擬環境中，TA-AFO與TFO及Elixir相較之下，平均轉發速率最高可分別提升約18% 及10%，而Drop Rate則可降低至94% 及 93%，並且能夠更快地對網路流量的變化進行反應。;Nowadays, cloud data centers integrate technologies such as Software-Defined Networking (SDN), Network Function Virtualization (NFV), and Hardware Offloading to strengthen the network architecture and accelerate packet processing through programmable networking, allowing the network infrastructure to handle more traffic and maintain more stable operations. Open vSwitch (OVS) is a common virtual switch in SDN/NFV architectures that utilizes SmartNICs to offload forwarding rules to hardware forwarding tables, significantly accelerating packet forwarding efficiency. However, the limited space in hardware forwarding tables necessitates the development of suitable offloading strategies to maximize their utilization. Most research typically focuses on offloading and replacing all suitable forwarding rules up to a fixed limit, without the capability to dynamically adjust the required hardware forwarding table space based on current network traffic changes. This study leverages the operational scenarios of data center networks to propose a Traffic-Aware Adaptive Flow Offloading (TA-AFO) strategy, integrating the software and hardware components of network virtualization architectures. TA-AFO continuously monitors traffic changes on OVS-DPDK and dynamically adjusts the thresholds for offloading forwarding rules to SmartNICs, thereby reducing hardware space consumption and ensuring forwarding efficiency. Furthermore, TA-AFO will be integrated into OVS-DPDK and validated using 200G Nvidia BlueField-2 SmartNICs. Experimental results show that, in a simulated environment with real network traffic, TA-AFO achieves an average forwarding rate improvement of up to approximately 18% and 10% compared to TFO and Elixir, respectively. Additionally, the Drop Rate is reduced by up to 94% and 93%, and TA-AFO responds more quickly to changes in network traffic.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	10	View/Open

社群 sharing

Loading...