Topological analysis of cloud service connectivity

摘要

Relying on services in the cloud involves manifold availability risks and concerns. This article focuses on the network reachability of cloud services. We present a study on cloud outages and causes, and analyze the topological connectivity of major cloud service providers (CSPs) by graph-based measures. Our approach is based on the construction and integration of an empirical dataset describing the connections between Autonomous Systems (ASs) of organizations that form the Internet backbone. According to our findings, though the ASs of CSPs generally appear to be better connected than an average AS, they also vastly differ in several connectivity measures, sometimes by more than an order of magnitude. Our results help to identify well-connected CSPs and CSPs that could potentially suffer more from Internet outages, if no additional path redundancy is provided. Our approach can be used by CSPs to assess connectivity beyond their own premises. It can also support cloud service customers during benchmarking and selection of CSPs when high availability is a critical requirement.