.ISM File Format – Live ingest

Smooth Streaming server manifest

For a live presentation there is no file on disk to start with. Each encoder sends its part of the server manifest to the server at the start of a long-running HTTP POST, followed by the media. MS-SSTR section 2.2.7 specifies this.

The publishing point address

http://server/live/channel1.isml/Events(evening)/Streams(720p)
The presentation path ends .isml. Events(…) is optional; Streams(…) names this connection’s stream.
PartMeaning
http or httpsProtocol
path.ismlThe publishing point
Events(id)Optional. Lets one address be reused for a new event without stale copies in caches. Every encoder must use the same one (default empty).
Streams(id)Identifies the stream, such as Streams(720p). Encoders that send the same stream ID to separate addresses act as backups; the server drops duplicate and out-of-order fragments.

The POST body

Part, in orderContent
ftyp boxMajor brand isml, minor version 1, compatible brands including piff and iso2
uuid box: StreamManifestBox (optional)UUID 3C2FE51B-EFEE-40A3-AE81-5300199DC348, a version, 3 flag bytes, then a SMIL document listing every stream the server must wait for (below)
uuid box: LiveServerManifestBoxUUID A5D40B30-E814-11DD-BA2F-0800200C9A66, a version, 3 flag bytes, then the live server manifest: a SMIL document like an .ism (see Server manifest elements)
moov boxThe track headers of the fragmented MP4 stream
fragmentsmoof and mdat pairs, as long as the broadcast lasts

Both SMIL documents are XML 1.0 without a DTD, in the SMIL 2.0 namespace. The live server manifest has one meta in its head, and every track’s src is Streams.

The stream manifest

If present, the server does not start the broadcast until every stream listed has connected. Leave it out to start as soon as the first encoder connects.

<smil xmlns="http://www.w3.org/2001/SMIL20/Language">
  <body>
    <par>
      <ref src="Streams(720p)"/>
      <ref src="Streams(360p)"/>
    </par>
  </body>
</smil>
The server waits for two streams, each from its own encoder.

Worked example

<smil xmlns="http://www.w3.org/2001/SMIL20/Language">
  <head>
    <meta name="title" content="Channel1"/>
  </head>
  <body>
    <switch>
      <video src="Streams" systemBitrate="2500000">
        <param name="trackID" value="1" valuetype="data"/>
        <param name="trackName" value="video" valuetype="data"/>
      </video>
      <audio src="Streams" systemBitrate="128000">
        <param name="trackID" value="2" valuetype="data"/>
        <param name="trackName" value="audio" valuetype="data"/>
      </audio>
    </switch>
  </body>
</smil>
A live server manifest for one video and one audio track, as an encoder sends it inside the LiveServerManifestBox.