What the race measures
- First frame: time from pressing start until the video plays. For Slab, this includes finding the shard holders, fetching the manifest and decoding the first slabs.
- Stalls: how often playback stopped to wait for data, and for how long in total. 1080p here needs about 2.3 Mbit/s, sustained.
- Seek: jump to 5:00 and time how long until it plays again. Every network has to fetch from the middle of the file (an HTTP range read).
- Website: time until each copy of this website has finished loading in its frame: HTML, CSS, fonts and images, each a separate request.
How to read the result
- Location matters. All of Slab’s testnet nodes are in one data centre in Germany. Swarm and IPFS run worldwide; the S3 bucket sits in one region. Readers far from Germany are at a disadvantage on Slab, readers near it at an advantage.
- Caches matter. Slab requests bypass the CDN in front of slab.me.uk (each run adds a unique query string), so every Slab byte comes from the storage nodes. Swarm and IPFS gateways keep what earlier visitors loaded, so the first viewer of the day waits longest. S3 is served straight from its bucket, without a CDN. Measured cold reads of Slab and Swarm, 1 KB to 50 MB, are on Compare.
- The same bytes everywhere. Each network holds the same MP4 file (index at the front, so playback starts before the download ends) and the same build of this website.
Big Buck Bunny © Blender Foundation, peach.blender.org, licensed under CC BY 3.0. Re-encoded at 1080p, 2.2 Mbit/s.