A scalable architecture for federated service chaining

The orchestration of Service Function Chain in multiple clouds calls for low-cost, low-latency and scalability. In the existing literature, several techniques have been proposed to meet these requirements. However, how to federate service chains across geo-distributed clouds in light of these requirements remains open. This thesis aims to study how to compose service chains across multiple clouds by considering several factors such as domain autonomy, domain confidential information, scalability, deployment cost, end-to-end latency and dynamic traffic demands. In particular, the proposed schemes in this thesis are devised to improve the most crucial performance metrics: the deployment cost and the end-to-end latency. First, we propose a distributed architecture that jointly considers domain autonomy, domain confidential information and scalability. This architecture enables service chains across multiple administrative domains without revealing sensitive network information such as the domain topology. The proposed architecture significantly reduces the deployment cost which consists of resource and traffic routing costs. Moreover, the proposed architecture remarkably reduces the execution time which suggests that it processes the SFC requests timely. Second, the network traffic is dynamic in nature. To accommodate the varying traffic demand in edge clouds, it is important to dynamically scale VNFs in an agile and efficient manner by considering the resource scarcity at the edge. Hence, we propose a bottleneck-aware VNF scaling and traffic routing algorithm to effectively handle the incoming traffic. The proposed algorithm uses vertical and horizontal scaling in light of the VNF category. The experimental results show that the proposed algorithm efficiently shortens the end-to-end latency, improves the VNF utilization rate and reduces the running time.