The ability to tolerate, remediate, and recover from network incidents (e.g., caused by device failures and fiber cuts) is critical for building and operating highly-available web services. Achieving fault tolerance and failure preparedness requires system architects, software developers, and site operators to have a deep understanding of network reliability at scale, along with its implications to data center systems. Unfortunately, little has been reported on the reliability characteristics of large-scale data center network infrastructure, let alone its impact on the availability of services powered by software running on that network infrastructure (service-level availability).
This paper fills the gap by presenting a large-scale, longitudinal study of data center network reliability based on operational data collected from the production network infrastructure at Facebook, one of the largest web service providers in the world. The study covers reliability characteristics of both intra and inter data center networks. For intra data center networks, we study seven years of operation data comprising thousands of network incidents across two different data center network designs – a classic cluster-based architecture and a state-of-the-art fabric-based topology. For inter data center networks, we study eighteen months of recent repair tickets in the field to understand reliability of WAN backbones. In contrast to prior work, we study the effects of network reliability on web services, and how these reliability characteristics evolve over time. We discuss the implications of network reliability on the design, implementation, and operation of large-scale data center systems and how it affects highly-available web services. We hope our study forms the foundation of understanding the reliability of large-scale network infrastructure, and inspires new reliability solutions to network incidents.