Maximizing Encoding Speed with Windows Azure Media Service Encoding

Question

I have an android application (client), asp.net web api web server (server), and Windows Azure Media Services (WAMS) account.

What I Want: To upload a 3-30 second video from the client to the server and have it encoded with WAMS then available for streaming via HLSv3 as quickly as possible. Ideally a video preview image would be generated as well. As fast as possible is something like sub one minute turn around. That's likely not realistic, I realize, but the faster the better.

Where I'm At: We upload the video to the server as a stream, which then stores it in Azure blob storage. The server returns to the client indicating upload success. The server has an action that kicks off the encoding which then get's called. I run a custom encoding task based off of the H264 Adaptive Bitrate MP4 Set 720p preset modified for taking a 640x480 video and cropping it to 480x480 at the same time as encoding. Then I run a thumbnail job that generates one thumbnail at 480x480. Depending on the reserved encoder quality this can take ~5 mins to ~2 mins. The encoding job time is only 30-60 seconds of that and the rest is a mix of queue time, publishing time, and communication delay.

What can I do to improve the client upload to video streamable turn around time? Where are the bottle necks in the encoding process? Is there a reasonable max speed that can be achieved? Are there config settings that can be tweaked to improve the process performance?

Mike Goatly Mike Goatly · Accepted Answer · 2014-10-31T15:24:41

Reduce the number of jobs

The first thing that springs to mind is given you're only interested in a single thumbnail, you should be able to consolidate your encode and thumbnail jobs by adding something like this to the MediaFile element of your encode preset:

  <MediaFile ThumbnailTime="00:00:00"
    ThumbnailMode="BestFrame"
    ThumbnailJpegCompression="75"
    ThumbnailCodec="Jpeg"
    ThumbnailSize="480, 480"
    ThumbnailEmbed="False">

The thumbnail will end up in the output asset along with your video streams.

Reduce the number of presets in the task

Another thing to consider is that the preset that you linked to has multiple presets defined within it in order to produce audio streams at different bitrates. My current understanding is that each of these presets is processed sequentially by the encode unit.

The first preset defines the video streams, and also specifies that each video stream should have the audio muxed in at 96kbps. This means that your video files will be larger than they probably need to be, and some time will be taken up in the muxing process.

The second and third presets just define the audio streams to output - these wouldn't contain any video. The first of these outputs the audio at 96kbps, the second at 56kbps.

Assuming you're happy with a fixed audio quality of 96kbps, I would suggest removing the audio from the video streams and the last of the audio streams (56kbps) - that would save the same audio stream being encoded twice, and audio being muxed in with the video. (Given what I can tell from your usage, you probably don't need that anyway)

The side benefit of this would be that your encoder output file size will go down marginally, and hence the cost of encodes will too.

Workflow optimisation

The only other point I would is regarding the workflow by which you get your video files into Azure in the first place. You say that you're uploading them into blob storage - I assume that you're subsequently copying them into an AMS asset so they can be configured as inputs for the job. If that's right, you may save a bit of time by uploading directly into an asset.

Hope that helps, and good luck!

Maximizing Encoding Speed with Windows Azure Media Service Encoding

1 Answers