Distributed Adaptation Framework
In today’s
multimedia content delivery architecture, adaptation becomes more and more
important. Content providers aspire towards serving a plethora of heterogeneous
end devices and networks without neglecting economical principles, i.e., if
multiple versions of the same content are maintained. Therefore, a single high
quality multimedia resource is stored on the server and adapted according to the
usage environment on demand. The MPEG-21 Digital Item Adaptation (DIA) standard
specifies normative description tools enabling the construction of device and
coding-format independent adaptation engines in an interoperable way. However,
it is not realistic that a single adaptation node (or module) could cope with
all kinds of usage environments. As a consequence, different adaptation nodes
distributed over the whole network can be employed, specifically for serving
different access networks. Interoperability among these nodes can be guaranteed
through standardized media and metadata formats. This requires that the metadata
associated with the multimedia content needs to be transported to such
adaptation nodes in order to steer the actual adaptation process
there.
DIA enables generic adaptation of media
resources. This means, that the adaptation engine does not need any knowledge
about the media it is about to adapt, all information which is necessary is
provided from the outside. This information (called “Tools”) is standardized in
the DIA standard. One important Tool in order to understand DIA is the (generic)
Bitstream Syntax Description (gBSD). This tool provides a high level XML
description of the media.
Given this gBSD and a scalable resource, in short the adaptation happens as
follows:
- The gBSD is transformed based on
external constraints (such as the available bandwidth)
- Based on this transformed gBSD
specific parts of the resource are thrown away (thus the need for a scalable
resource)
In order to perform DIA all Tools
(XML descriptions) are needed in order to provide all the necessary information
(bitstream description, constraint descriptions, …). Thus, to enable distributed
adaptation it is necessary to transport these Tools to any node which would like
to perform DIA. As the Tools vary in complexity and size, this is a non-trivial
problem. Efficient solutions to this problem are therefore one of the key
aspects of the DANAE Distributed Adaptation Framework whose current architecture
is depicted above.
The constraint descriptions
(e.g. the available bandwidth) are made available to the Adaptation Node through
the context collection chain. The gBSD and other content-related metadata is
streamed in synchronization with the media data in order to support distributed,
generic adaptation in streaming scenarios. In order to support multicast and
broadcast scenarios, random access into the metadata stream needs to be
possible. Therefore, the metadata is first transformed into fragments which can
be processed independently (and thus provide random access points - RAPs), as
depicted above. Afterwards, the fragments (called Process Units) are encoded
using the MPEG-21 Binary XML (BiM) codec, which enables to decide whether to
stream the complete fragment (RAP) or only the changed information compared to
the last fragment (depicted below).
| 
