Recently, I helped a few short video teams to deal with the problem of lagging, and found that eight out of ten died on the "thought that everything would be fine if they bought a CDN". As a result, once on line, the player rotates more diligently than the electric fan, the user turnover rate directly soared to 30%. these days, even the CDN has to "prevent teammates" - the acceleration you think may not be used in the right place at all.
Slow loading of short videos is never as simple as insufficient bandwidth. I took the flame map to pick up a few cases, found a triple whammy: the first is the source station was DDoS penetration, CDN back to the source of the collateral paralyzed; the second is the video encoding is rough, 2 minutes of video hard to press out the volume of 200MB; the third is the edge of the node caching strategy is lousy, the user every time to download the same content over and over again. The most pitiful is the "intelligent compression" of some vendors, actually 1080p video compression out of the mosaic texture, this wave of operation is really negative optimization.
First of all, how to choose a high-defense CDN. Don't believe those "unlimited protection" nonsense, tested CDN5 T-level cleaning capacity can really stand up to SYN Flood mixed CC attacks, but 08Host billing by volume encountered unexpected traffic is simply burning money like water. The key is the hidden skills: any CDN to open the "cache penetration protection", or hackers directly penetrate the cache to hit the source station, buy more protection is useless. Configuration example see here:
Video compression is the real bandwidth killer. I've seen teams use H.264 to compress 4K video and open CRF 18, and a single video eating 3GB of traffic is purely suicidal behavior. Tested HEVC + HDR combination of file size can be compressed down 40% without loss of clarity, but pay attention to compatibility - some old cell phones decoding HEVC can directly let the CPU soared to 90 degrees.
My pressure parameter scheme has been verified in over three hundred videos:
Edge caching strategy is more complicated than expected. CDN07's intelligent warm-up function is worth boasting, which can cache to edge nodes in advance according to the list of popular videos. However, the TTL settings of most vendors are anti-human: they actually set a 1-hour cache for the m3u8 index file, which is obviously the right way to use a 5-second short cache with ETag authentication. Remember a principle: static slices are cached for 30 days, dynamic lists are cached for 5 seconds, and key files are never cached.
The real secret to speeding up is in the HTTP/2+Brotli combo. Test data shows that Brotli compresses m3u8 files saving another 17% volume over gzip, especially for those playlists that contain a lot of timestamps. But you need to manually turn on the support in the CDN background, the default settings are basically off:
Lastly, let's dump a real test data: after a short drama app with one million DAU accessed the hybrid solution - using CDN5 for high defense resolution + 08Host to handle overseas traffic + self-built edge caching layer - the first frame loading time was pressed down from 4.3 seconds to 1.1 seconds, and the monthly bandwidth cost was directly reduced by $10,000,000 per month. Monthly bandwidth costs dropped by 620,000 dollars. The key work is in the details: even the video slice size is re-adjusted to 2 seconds per section, so that the seek operation can reduce the number of requests of 30%.
Now look at those who are still using the "one-click acceleration" program team, it is like looking at a person with a scalpel to cut vegetables - not the tool does not work, but the use of the road is not right at all. Short video acceleration is a systematic project, from encoding to transmission to caching, each link must be a technical person's more serious vigor to die. After all, the speed of the user's finger slide away, much faster than your video buffer speed.
