BiliBoost

Watching Bilibili from overseas hits the same wall every time: bandwidth is fine, but the video won’t load. A gigabit connection can’t push 1080p smoothly. It plays for a few seconds, rebuffers, or the progress bar just sits there. Popular videos work because they’re already in the edge cache. Unpopular ones miss it, both the primary and backup CDN nodes time out, and you’re stuck.

GitHub has plenty of CDN-switching scripts that try to fix this. But a single cross-border connection has a hard ceiling on throughput, so for high-bitrate video, swapping nodes barely moves the needle.

BiliBoost raises that ceiling. It aggregates buffering across multiple connections and picks better CDN nodes so the aggregation has good material to work with.

What it is

A single JavaScript file, run through Tampermonkey, that does both jobs inside the browser.

Node selection

Candidate nodes are probed with two-point sampling: the file head plus a deep offset into the file. A node that only cached the first few megabytes looks fast on a head-only probe, then stalls mid-video. Sampling a distant range catches those false positives before the player commits to a node.

Beating the per-connection throttle

The throttle is per connection, so the script splits large media requests into byte ranges, fetches from up to four nodes simultaneously, and stitches the pieces back into one stream with proper backpressure. In my measurements, a link that managed about 2 Mbps on a single stream aggregates to 22–103 Mbps.

Latency gets the same treatment. If a node hasn’t answered within about 900 ms, a second request goes to a backup node and the two race. First response wins.

Cache misses

When a node 404s or stalls, playback falls back down the candidate chain while better nodes are warmed in the background for the next segment. Under fairly extreme conditions, cold content that used to take 37 seconds to start now starts in about 250 ms. 4K plays fine.

Caveat

Tested on a network in California, US West Coast. The CDN selection strategy may work best there; other regions haven’t been tested.