Vector Netflix Performance Monitoring

Vector is an open-source host-level performance monitoring framework originally developed at Netflix to provide on-demand, real-time system performance monitoring across large-scale infrastructure. Created by the Netflix Performance and Reliability Engineering team (Los Gatos, California) and released as open source in 2015, Vector uses the Performance Co-Pilot (PCP) framework as its backend for collecting and exposing system-level performance metrics. Key features: on-demand monitoring: unlike traditional always-on monitoring agents, Vector allows engineers to enable monitoring on any host instantly for real-time debugging and performance analysis, making it ideal for troubleshooting production issues. Metrics collection: uses Performance Co-Pilot (PCP), a system-level performance monitoring toolkit originally from Silicon Graphics (SGI), to collect hundreds of system and application metrics including CPU utilization (per-core, per-process), memory usage, disk I/O (per-device throughput, latency, queue depth), network (interface traffic, TCP connections, retransmissions), process statistics, filesystem usage, NFS, and interrupt handling. Web UI: a real-time dashboard built with AngularJS that displays live performance graphs with customizable time windows (1 minute to 24 hours), supporting multiple metric overlays, zoom, and comparison. Metric browser: search and select from hundreds of available metrics with auto-complete. Custom dashboards: save and share dashboard configurations. Architecture: Vector runs as a lightweight web application that connects to a local or remote PCP pmcd daemon. Agentless monitoring: for network devices and remote hosts, Vector can connect to remote PCP daemons via TCP. Developed at Netflix for managing their AWS infrastructure serving 200+ million subscribers. Cross-platform: Linux, macOS, Windows (via PCP). Apache-2.0.