Link/Switch Failure Analysis of Data Center Networks on Matroidal Connectivity

With the surge of bandwidth demand for cloud applications and the exponential growth of data, data center networks (DCNs) are expanding rapidly, followed by the daily increasing likelihood of failures. Such failures, whether due to device or link issues, are inevitable and often lead to packet loss, transmission delays, and even system downtime. Thus, it is crucial to assess the fault-tolerant capabilities of data center networks using appropriate reliability metrics when failures occur. BCube is a well-known server-centric data center network with many advantages, such as rich low-diameter paths, high throughput, and excellent expandability. Not only do the recently proposed matroidal connectivity and conditional matroidal connectivity have reasonable fault assumptions that align well with the structural characteristics of data center networks, but they also significantly enhance the fault tolerance performance of DCNs. This paper determines the matroidal connectivity and conditional matroidal connectivity of BCube, which is the first study to apply the two reliability metrics in DCNs. Then, we extend the conclusions about (conditional) matroidal connectivity to analyze the fault tolerance of BCube in the occurrence of switch failures. In addition, we develop an efficient algorithm to identify the structural features of minimum faulty edge sets, where the cardinality of these edge sets corresponds to the conditional matroidal connectivity of BCube. Finally, we experimentally evaluate the effects of both link and switch failures on BCube’s performance under the matroidal restriction. The experimental analyses reveal that BCube DCNs exhibit high fault tolerance under matroidal constraints, with the ability to withstand both link and switch failures.