The Service Analysis and Network Diagnosis DataPipeline

Abstract: Modern network performance monitoring toolkits, such as perfSONAR, take a remarkable number of measurements about the local network environment. To gain a complete picture of network performance, however, one needs to aggregate data across a large number of endpoints. The Service Analysis and Network Diagnosis (SAND) data pipeline collects data from diverse sources and ingests these measurements into a message bus. The message bus allows the project to send the data to multiple consumers, including a tape archive, an Elasticsearch database, and a peer infrastructure at CERN. In this paper, we explain the architecture and evolution of the SAND data pipeline, the scale of the resulting dataset, and how it supports a wide variety of network analysis applications.