Large-scale Measurement and Analysis of Cloud Service Reliability

Given the growth and our increasing reliance on cloud services, it is crucial that we have a better understanding of the reliability of cloud services. Today, a cloud service is composed of a complex array of building blocks and any of the building blocks can fail. Prior empirical studies on cloud reliability have mostly focused on a limited set of cloud services and a few building blocks, and fail to capture a broader view of the reliability of cloud services. Overcoming the above limitations requires data about failure incidents in a large number of diverse cloud services. We leverage the fact that today's cloud services publicly expose detailed information about failure incidents on the web. By crawling such publicly available incident reports, we conduct the first large-scale measurement study of the reliability of 152 cloud services, covering over 11,000 incidents over a period of up to 3 years. Our analysis of a diverse set of cloud services provides a better understanding of their failure rates, the different types of failures they experience, their recovery time from failures, and their availability. We also investigate other factors that may contribute to a service's reliability, such as maintenance, provider faults and security incidents.