Meta Research

Datacenters are characterized by their large scale, comprising a large number of network links and switches. However, these hardware components can develop intermittent faults, resulting in randomly occurring packet drops or delays that harm application performance—several such faults occur daily in large production datacenters. Since the effects are intermittent, traditional detection techniques involving end-host and router statistics or active probe traffic can fall short in their ability to identify and locate these errors. In this article, we present our passive hybrid approach that combines network path information with end-host-based statistics to rapidly detect and pinpoint the location of datacenter network faults inside a production Facebook datacenter.