虛擬機器監視器和容器技術應用於虛擬EPC之效能評估

Abstract

隨著IoT、AR/VR等新興技術的發展，現行4G行動網路的高成本、低彈性、限制創新等問題相繼被提出，使用SDN與NFV技術將有助於4G行動網路提升彈性、開放創新並同時降低網路建置與維運成本。然而將虛擬化技術應用於行動網路之效能評估的相關研究不多，大多數的研究傾向於單一機器設備上運行虛擬化環境後做一般性的效能測試，其效能評估中都顯示容器與實體機的網路效能有較大的落差。

然而，網路效能為行動網路效能中一項重要的指標，為了補足虛擬化EPC之效能評估相關研究的不足，本篇論文探討了包括實體機、KVM、Docker等共四種網路環境下的效能表現，並在基於SDN概念設計的行動網路 - Open5GCore之上對用戶層效能與控制層做效能評估。實驗證明就Open5GCore行動網路而言，Docker的網路效能接近於實體機，KVM_VirtIO約有10%的效能損失，KVM_rtl8139則最多有27%的效能損失，造成效能損失的原因為不同虛擬化環境有著不同的I/O路徑以及系統架構，相關的實驗數據及結果分析可供網路服務供應商參考。

People bring up several shortcomings of 4G mobile network such as high cost, low flexibility, and restricted innovations while the Internet-of-things and AR/VR becoming more and more popular. Software Defined Networking (SDN) and Network Function Virtualization are two popular emerging technologies that possible to solve such problems. However, there are few studies about evaluating virtualized EPC. Most of the studies are about evaluating on one device over virtualization platforms. The results show a gap of network I/O performance between container-based virtualization and native setup.

Network I/O performance is an important metric for the mobile network. This paper evaluates an SDN-based mobile network, Open5GCore, over four different network environments, includes the user plane and control plane tests. These results have shown that the performance of network I/O of Open5GCore for Docker is close to the native setup. There is about 10% overhead for KVM_VirtIO and at most 27% for KVM_rtl8139. The reason for such costs is due to the virtualization techniques have different I/O path and system architecture. The results and analysis are available for network providers.