4 ms·
How bitcodin uses Google Compute Engine to encode video 100x faster than others
- dr_zoidberg 11y agoI got the feeling that they were about to explain the details when they decided the article was long enough and that was it... They say their coding approach is not tied to GCE, but in the end GCE's infrastructure is great! Does anyone have some interesting details or a good read on distributed encoding? I imagine that since they have a segmented input, they can paralellize extremely well, encode every segment independently and in the end stitch everything together (or maybe not, and keep them separate to simplify streaming?).
- slederer 11y agoThanks, good point! Yes, the approach in general is not limited to GCE, but some characteristics of GCE - like the fast provisioning time as well as the good interconnection between instances - are very beneficial for that approach. That's the reason why it works so fast on GCE. Does this makes sense to you? Our input files are not segmented, you can use the commonly used media formats out in the market as MP4, MKV, etc. and we take care of the rest.
- dr_zoidberg 11y agoWell most video containers formats allow for some sort of segmenting that's also beneficial for streaming applications. My guess was that segmenting the video on the "natural chunks" of the format allowed for easier paralellization.
- slederer 11y agothat's completely true, containers like ISO Base Media FF(MP4) or Matroska support streaming use cases, which is even more beneficial if we got those files as input. But we can also work with any others ... which was also pretty much work :-)