Efficient Approaches for Homing Complex Network Services

Abstract

Network service providers (NSPs) offer a wide array of network services to their customers at a global scale. In recent years, NSPs have been migrating their infrastructures to a virtualized software-based one enabled by the network functions virtualization (NFV) paradigm. One critical aspect in operating NFVbased network services is homing. Homing (or placement) of virtual network functions (VNFs) on cloud and network service provider (NSP) infrastructures is a crucial step in the orchestration of network services, involving complex interactions with the cloud, SDN and service controllers. Large NSPs process thousands of homing requests submitted by their customers on a daily basis. At such large scale, it is imperative that homing of network services is performed efficiently. In this dissertation, and guided by our extensive discussions and collaboration with a Tier-1 NSP, we identify limitations and challenges across multiple layers of the homing stack. Starting at the bottom of the stack, we look at how it is extremely challenging to provision VNFs in a truly elastic manner – hindering the ability to efficiently manage them. At upper layers, we identify limitations with current approaches that are used for service and cloud controllers to aggregate data from end nodes. We analyze why such approaches are not efficient when deployed at large scale. Finally, we identify several dependency problems that result from deploying distributed instances of the homing service. Accordingly, we design systems that efficiently address such challenges across the homing stack. Specifically, we design a novel stateless architecture for VNFs to provide true elasticity when deployed at NSP cloud sites – allowing VNFs to seamlessly scale and failover. In addition, we propose and design a peer-to-peer search service that offers real-time data retrieval at global scale in an efficient manner. We also design and evaluate a novel homing service that provides quality homing solutions while significantly reducing load on the service and cloud controllers. We extensively evaluate each of these solutions and demonstrate their efficiency in addressing the different challenges across the homing stack.