Towards a High-performance and Reliable System for Emerging Edge Network

5G is a new paradigm that enables tremendous opportunities by delivering high-bandwidth and low-latency network. The emerging 5G-enhanced applications including smart home, smart factory, Artificial Intelligence (AI), Augmented Reality (AR)/ Virtual Reality (VR) and autonomous driving are widespread through the deployment of the 5G. In responding to the fast-varying user service requirements and highly mixed traffics, Network Function Virtualization (NFV) technique is emphasized by the 5G to enhance its functional and architectural viability. NFV is a novel paradigm that packages the network services as virtual machines (VMs) or containers on Commercial-Off-The-Shelf (COTS) servers instead of traditional vendor proprietary hardware devices. In this dissertation, we focus our performance characterization and optimization on the emerging virtual Radio Access Network (vRAN) system enabled by the NFV technique, since it plays a vital role in today’s edge infrastructure for its better support for the latency-sensitive applications. The vRAN has become an essential infrastructure to deploy the emerging edge applications, especially in the new-coming Infrastructure-Augmented Autonomous Driving (IAAD) system. This dissertation sets to illustrate the key challenges and their corresponding solutions of the vRAN’s infrastructure/architecture, mainly focusing on the vRAN’s Single Instruction Multiple Data (SIMD) mechanism and its management/control layer, to match the deployment requirements of the emerging edge applications from the performance and reliability perspective. To guarantee the performance and the reliability of the emerging edge applications, the vRAN edge network infrastructure must possess the ultra-low latency feature which is scarcely achievable by the traditional network infrastructure. However, the vRAN edge network infrastructure, specifically the Commercial-Off-The-Shelf (COTS) servers, has limited computing resources, which will hinder the vRAN edge network to meet the latency requirement applied by the emerging edge applications. Moreover, the vRAN edge applications have fundamentally differentiated computing algorithms which will ignite various resource utilization patterns. Consequently, there exists ”Inefficient computing resource utilization” caused by the mismatch between emerging edge applications’ properties and the COTS micro-architectural structure, which will degrade the performance of the edge applications. Besides, the emerging edge applications, especially the IAAD system, demand ultra-high reliability provision for the vRAN edge system. However, the current vRAN can not satisfy the provisions requested by the new-coming IAAD system, which will expose the IAAD system to the severe safety issue. To tackle the challenges, we propose a SIMD consciousness computing mechanism and a data fusion awareness management methodology to thoroughly exploit the micro-architecture of the state-of-the-art COTS servers and entirely utilize the management layer of the current vRAN infrastructure. The proposed solutions can effectively promote the vRAN infrastructure’s performance and the reliability of the emerging IAAD system. Specifically, 1) We implement a thorough architectural characterization on the key network components and MEC applications on the vRAN edge system to provide guidance on the hardware architecture design trade-off of vRAN edge COTS servers. 2) We propose ”Arithmetic Ports Consciousness Mechanism” (APCM) to exploit the idle architecture resources to eliminate the ubiquitous backend bound of the current vRAN infrastructure and promote the throttled bandwidth between the COTS server’s registers and L1 cache. 3) We develop a new architecture between the COTS server’s registers and L1 cache to avoid the congestion caused by the data arrangement procedure of the vRAN system and the emanating AI applications, which will effectively accelerate the proceeding of the current edge infrastructure.4) We create a Spatial-Temporal (S-T) fusion layer above the current control layer to tackle the edge network fluctuation challenges of the current vRAN infrastructure to guarantee the reliability requirement applied by the emerging IAAD system. This dissertation aims at developing a high-performance and reliable system for the emerging edge network to deploy emerging AI-enable applications and IAAD system efficiently and effectively