Zero-Copy NVMe Sendfile
Large ONNX checkpoints, Parquet shards, and ZIM bundles stream directly from Linux kernel page cache over 16x concurrent aria2c or rclone ranges without user-space buffer copies.
Public distribution node in Los Angeles serving multi-gigabyte ML model shards, reproducible compiler toolchains, ZIM knowledge archives, and persistent HTTP/2 Byte-Range PMTiles streams over NVMe page-cache rings.
Combined bulk archive egress (Gbps) and multiplexed sparse-range request rate (req/s). Hover to inspect intervals.
Click any row to open the checksum inspector and copy multi-connection CLI commands. Press / to search.
| Artifact | Category | Size | Access Workload | 24h Pulls | Action |
|---|
Engineered specifically for two contrasting public mirror traffic profiles on shared TLS 1.3 endpoints.
Large ONNX checkpoints, Parquet shards, and ZIM bundles stream directly from Linux kernel page cache over 16x concurrent aria2c or rclone ranges without user-space buffer copies.
Package lockfile resolvers and build agents keep persistent TLS 1.3 sessions open, dispatching hundreds of 4 KiB to 64 KiB metadata and hash manifests per second over a single socket.
Planetary OpenStreetMap vector archives respond to Range: bytes=offset-end queries in sub-millisecond edge time, turning one static archive into a stateless map backend.