CDN Request Traces for Video Streaming Research

Two anonymized, request-level CDN traces from two production video-on-demand streaming platforms, released as companion datasets to the paper "A Production Trace From a Public CDN for Caching and Content-Placement Research" (IEEE ICNP, CEC workshop, 2026). Both traces record one row per HTTP segment request and include the production edge-cache decision (hit/miss), opaque object identity (video, segment, rendition), response size, HTTP status, and opaque client and infrastructure identifiers, under a shared anonymization scheme with aligned column names. No personal data is included. Dataset A (dataset_a.csv.gz, 148 MB compressed, 8,175,737 rows): six months (2025-10-26 to 2026-04-24), 904 videos, two adaptive-bitrate protocols (DASH 86.7%, HLS 13.3%) with a DASH audio companion track, plus per-request timing fields (TTFB, transfer time, throughput, TLS overhead) enabling evaluation on a quality-of-experience axis. Dataset B (dataset_b.csv.gz, 918 MB compressed, 60,315,746 rows): fifteen days (2024-05-27 to 2024-06-10), 3,664 videos, six rendition tiers, edge-measured round-trip time, and fine-grained client ASNs (1,083 distinct). The traces support research on cache admission and eviction, prediction-based prefetching and cache warming, content placement, and workload characterization. Demand in both traces is highly concentrated (video-level Gini >= 0.88) with substantial day-over-day hot-set churn.