Abstract: With the large-scale increase of users, the reliability of the cloud system has become a challenging issue in the industry and academia. Many researchers have studied the reliability mechanism of cloud computing systems and proposed reliability awareness methods to achieve resource integration and improve system reliability. However, various hardware and software failures occur inevitably and cannot be accurately found and repaired in a timely manner. Moreover, since most of the studies cannot determine the background operation mechanism of the cloud system, this brings significant problems to the research of cloud computing reliability. To solve this problem, we first extract the key features that can be used to increase system reliability in cloud computing architectures. Secondly, we present an architecture framework for reliability simulation and analyze four types of common system failures: hardware failures, virtual machine failures, data inconsistency failures, and service timeout failures. Finally, experiments and verification based on a set of realistic configurations and operation runtimes are implemented as an extension of a well-known cloud simulation tool, CloudSim, to illustrate how these failures affect the reliability of cloud computing systems and how different resource scheduling algorithms handle these failures.

Key words: cloud computing, reliability simulation, failures, virtual machine