Evaluation of edge cloud service scenarios with application specific routing in the network

This thesis presents the evaluation of edge cloud service scenarios with Application Specific Routing (ASR) in the network. Edge cloud service is a computing service provided by decentralized servers which reside at the edge of the network and provide services to client devices in the local area. By being in close proximity to end users, there is a reduction in latency that may provide improved Quality of Experience (QoE). The edge cloud provides services to Internet of Things (IoT) or Cyber Physical Systems (CPS), such as, mobile phones, sensors, vehicles and augmented reality using devices like Google glass, which host real time applications and require quick response. This work evaluates the performance of such edge cloud services for mobile users in the MobilityFirst (MF) architectural framework.;The Hadoop open source cloud software framework has been used in this study for parallel and distributed processing of jobs at the edge cloud site. The performance of Hadoop has been benchmarked for different network scenarios by varying the link quality: bandwidth and latency. Results show that both link parameters and edge cloud performance have a significant impact on the response time for a client's cloud service request. Hence, this work uses both these metrics in making routing decisions to improve end users QoE with help of Virtual Network (VN) capabilities. VN on MF enables specification of customized network topologies and routing algorithms for specific application scenarios such as cloud service. ASR is one such routing algorithm that makes path selection based on application specific parameters that affect end users' QoE such as the link and edge cloud performance metrics. The link performance metric incorporates bandwidth and latency in the link from client to the edge cloud. The edge cloud performance metric on the other hand captures the average job completion time at the edge cloud site which gives a sense of the workload, the number of nodes and the processing capability of the cluster. With the aid of ASR on MF VN, the best edge cloud site for a client's request is chosen with the objective to minimizing the overall response time.;Evaluation of users' QoE for edge cloud services was performed on the ORBIT testbed by replicating cloud services on multiple edge cloud sites. Workload was induced at each of these sites to control the job completion time and link quality is varied to affect the link performance. Response time for a client's request is evaluated using ASR and also using the basic anycast service in MF, for performance comparison. While anycast always chooses the nearest edge cloud, ASR provides the best response time by either choosing a lightly loaded cluster or by choosing a high quality link or a combination of both, and therefore provides a better QoE to the end user.