Great, thanks for being on top of the VP9 issue!
-Andrew
-Andrew Lih Associate professor of journalism, American University Email: andrew@andrewlih.com WEB: http://www.andrewlih.com BOOK: The Wikipedia Revolution: http://www.wikipediarevolution.com PROJECT: Wiki Makes Video http://en.wikipedia.org/wiki/Wikipedia:WikiProject_Wiki_Makes_Video
On Sat, Aug 15, 2015 at 2:43 AM, Brion Vibber bvibber@wikimedia.org wrote:
Just a quick note -- there's been some ongoing work in preparation for supporting WebM with the newer VP9 codec:
- fixing up our server configuration with updated ffmpeg & ffmpeg2theora
packages
- fixing up the player to support VP9 as well as VP8
- adding support for producing WebM transcodes in VP9 in addition to VP8
Big thanks to a bunch of folks over on the ops side who have been helping with the packaging! We had to backport some packages to get VP9 working, and ffmpeg2theora needed some patches to work correctly in newer versions.
At some point in the coming weeks, the video transcoding servers will be re-imaged with an updated operating system https://phabricator.wikimedia.org/T104747 (Ubuntu Trusty plus our custom packages). Once this is complete, uploaded WebM VP9 files should start producing Ogg Theora and WebM VP8 transcodes (currently you can upload them but they fail to work).
With TMH updates going out some time next week, playback of VP9 originals https://phabricator.wikimedia.org/T62272 should also start working in Chrome and Firefox.
These updates also improve support of Opus audio files https://phabricator.wikimedia.org/T42193, both standalone in Ogg audio and in WebM videos.
Once we're satisfied everything works, we'll consider enabling production of VP9 transcodes https://phabricator.wikimedia.org/T63805. These will be about the same quality as VP8, but use only half the bandwidth. Later when we figure out how to rig up adaptive streaming, this'll mean bandwidth-constrained clients can bump up one resolution step in VP9 versus VP8...
One big warning -- VP9 encoding is 2-4x slower than VP8 encoding at the same resolution... so we'll have to investigate comparative CPU load and whether 1080p transcodes actually complete reliably before timeouts etc hit. This may be something we have to roll out later after more testing.
But even if it takes us time to get VP9 output working, having VP9 input functional will fix some existing files, make it easier to import from YouTube and other sources that are more aggressively using VP9, and generally makes us more future-proof.
-- brion
Wikivideo-l mailing list Wikivideo-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikivideo-l