Network Working Group G. Deen
Internet-Draft NBCUniversal
Intended status: Informational L. Daigle
Expires: April 27, 2017 Thinking Cat Enterprises LLC
October 24, 2016
Glass to Glass Internet Ecosystem Introduction
draft-deen-daigle-ggie-02
Abstract
This document introduces the Glass to Glass Internet Ecosystem
(GGIE). GGIE's purpose is to improve how the Internet is used create
and consume video, both amateur and professional, reflecting that the
line between amateur and professional video technology is
increasingly blurred. Glass to Glass refers to the entire video
ecosystem, from the camera lens to the viewing screen. As the name
implies, GGIE's scope is the entire video ecosystem from capture,
through the steps of editing, packaging, distributed and searching,
and finally viewing. GGIE is not a complete end to end architecture
or solution, it provides foundational elements that can serve as
building blocks for new Internet video innovation.
This is a companion effort to the GGIE W3C Taskforce in the W3C Web
and TV Interest Group.
This document is being discussed on the ggie@ietf.org mailing list.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on April 27, 2017.
Deen & Daigle Expires April 27, 2017 [Page 1]
Internet-Draft GGIE Intro October 2016
Copyright Notice
Copyright (c) 2016 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. Motivation: Video is filling up the pipes . . . . . . . . . . 4
4. Video is different . . . . . . . . . . . . . . . . . . . . . 5
5. Historical Approaches to supporting Video on the Internet . . 6
5.1. Video as an application . . . . . . . . . . . . . . . . . 6
5.2. Video as a network problem . . . . . . . . . . . . . . . 7
5.3. Video Ecosystem Encapsulation . . . . . . . . . . . . . . 7
6. Problem Statement and Solution Criteria . . . . . . . . . . . 8
7. The Glass to Glass Internet Ecosystem: GGIE . . . . . . . . . 8
7.1. Related work: W3C GGIE Taskforce . . . . . . . . . . . . 9
8. GGIE work of relevance to the IETF . . . . . . . . . . . . . 9
8.1. Affected IETF work areas . . . . . . . . . . . . . . . . 9
8.2. Example use cases . . . . . . . . . . . . . . . . . . . . 9
8.3. Core GGIE elements . . . . . . . . . . . . . . . . . . . 11
9. Conclusion and Next Steps . . . . . . . . . . . . . . . . . . 15
10. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 15
11. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 15
12. Security Considerations . . . . . . . . . . . . . . . . . . . 15
12.1. Privacy Concerns . . . . . . . . . . . . . . . . . . . . 15
13. Normative References . . . . . . . . . . . . . . . . . . . . 16
Appendix A. Overview of the details of the video lifecycle . . . 16
A.1. Media Lifecycle . . . . . . . . . . . . . . . . . . . . . 16
A.2. Video is not like other Internet data . . . . . . . . . . 19
A.3. Video Transport . . . . . . . . . . . . . . . . . . . . . 21
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 21
Deen & Daigle Expires April 27, 2017 [Page 2]
Internet-Draft GGIE Intro October 2016
1. Introduction
In terms of shear bandwidth, the Internet's largest use, without any
close second competitor, is video. This is thanks to the
proliferation of Internet connected devices capable of capturing and/
or watching streamed video. As of 2015 there are reports that
YouTube users upload over 500 hours of video every minute, and that
during evening hours NetFlix accounts for a staggering 50+% of
Internet traffic. The number of users using the Internet for both
ends of the video create-view lifecycle grows daily worldwide, and
this is creating an enormous strain on the underlying Internet
infrastructure at nearly every point from the core to the edge.
While video is one of the most conceptually simple uses of the
Internet, it is perhaps one of the most complex technically, built
from standards created by a large number of organizations and groups
some dating from before the modern Internet even existed. Many
critical parts of this complex ecosystem were not created with either
video's particular characteristics or vast scale of popularity in
mind. This has lead to both the degradation of the viewer experience
and many Internet policy issues around access to bandwidth for video
and the needed infrastructure to support the continued explosion in
video transport on the Internet.
The pace of video growth has been faster than new bandwidth for the
past many years, and all indicators are that, instead of abating, it
is actually accelerating as new users, new ways of sharing video, and
new types of video continue to be added. The Cisco Visual Networking
Index an excellent source of detail on this subject.
The combined current high levels of bandwidth consumed by video, plus
the accelerating pace of video's growth mean that to meet users'
demand for video, we must do more than simply rely on adding more
bandwidth. While other traditional improvements such as more
efficient codecs with better compression ratios are expected to
contribute to keep video flowing on the Internet, many in the
Internet video technology world have explored options to see if any
new approaches could be added to the mix to help the problem. That
was the motivation behind the creation of the GGIE Taskforce within
the W3C in 2014 with the charter to examine the end to end video
ecosystem and identify new areas of opportunity to improve video's
use of the Internet.
The W3C GGIE taskforce explored ways that video uses the Internet and
developed a series of use cases detailing specific scenarios ranging
from video capture, the editing and production cycle, through to
delivery to viewers. Out of these use cases there emerged a
recognition that there might be a new opportunity to improve Internet
Deen & Daigle Expires April 27, 2017 [Page 3]
Internet-Draft GGIE Intro October 2016
video by enabling edge devices, and the underlying network to more
actively participate in making delivery optimization choices beyond
the simple ways the do currently.
The GGIE approach is to apply and evolve existing technologies to the
task of optimizing Internet video transport to permit applications,
video devices, and the network to more actively participate in making
smart access and transport decisions. This approach recognizes that
there are already extensively-deployed video infrastructure elements
that need to continue to work and be part of the optimized video
ecosystem. These deployed devices, applications, players, and tools
are responsible for the already high levels of video bandwidth
consumption, and to only address new devices would not be solving the
larger, most important problem. This is why GGIE is an evolution of
how video uses the Internet, and not a revolution involving wholesale
replacement of existing architecture or protocols.
GGIE is not a complete solution to the video problem. It provides
foundational building blocks that are intended to be used by
innovators in their work to create new optimizations, and novel
techniques to help address the video problem in the long term.
GGIE initially proposes a simple framework of three components that
will permit improved playback device identification of viewing
sources and enable network level awareness of video transport and new
cache selection chocies. GGIE proposes: Using existing content
identifiers as a means to identify a work, or title; Data level
identifiers to identify the encoded video data for a particular
manifestation of the title; A mapping service that permits bi-
directional resolution of these identifiers.
This document outlines the basic proposal for these three base GGIE
components and introduces the overall GGIE approach to evolving the
current video ecosystem by introducing basic standardized building
blocks for innovators to build upon the Glass to Glass Internet
Ecosystem.
2. Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in [RFC2119].
3. Motivation: Video is filling up the pipes
The growth in video bandwidth need is exceeding the growth in the
bandwidth provisioning. This trend is in fact accelerating, meaning
the growth rate of video is growing faster than the growth rate of
Deen & Daigle Expires April 27, 2017 [Page 4]
Internet-Draft GGIE Intro October 2016
provisioning. Traditional techniques of caching, higher efficiency
codecs, etc, are all being used to help address the probiem and have
helped the Internet to continue to support the growth of video thus
far.
Video has been the top use of Internet bandwidth for several years
and is larger than the bandwidth used by all other applications
combined. This trend is unlikely to ease or reverse itself as users
of the Internet continue to make Internet transported video one of
their top uses of the Internet, either for uploading and sharing
video they creator, or as a primary sources for viewing video to a
wide variety of viewing devices: computers, tablets, phones,
connected televisions, game consoles, and AV receivers.
Adding to user demand, video itself is continually experiencing
innovation introducing ever higher resolutions (SD, HD, 4K, 8K...),
higher video quality, new distribution services (live one to many
streaming), and new user uses. The Cisco Visual Networking Index
projects that by 2019 there will be nearly a million minutes of video
per second transported by the Internet, a making up 80-90 percent of
all IP traffic.
The movitation behind GGIE is to help find new methods that can be
brought to bear, in addition to all the exiting ones, to help manage
the explosion in Internet video.
4. Video is different
Video is different than other uses of the network due to its combimed
high bandwidth demands and high sensitivity to latency and dropped
packets. Streaming of basic high-definition 1080p requires bandwidth
in the low Mbps translating into Gigabytes for each hour of video,
all transported with consistent low latency and very little packet
loss in order to deliver a suitable watching experience the viewer.
This differentiates video from other Internet applications as some
have low latency and packet loss requirements but don't need high
bandwidth, while others may demand high bandwidth, they will tolerate
high latency and dropped packets. An email user can tolerate an
extra moment to retransmit dropped packets, and a web page user can
tolerate a slow DNS lookup, but a video viewer sees latency and
dropped packets as jittery playback and low bandwidth as a
fundamental barrier to streaming at all. From the user's perspective
the network has faield to meet their need. (Audio has similar
challenges in terms of intolerance of delay and jitter, but the data
sizes are significantly smaller).
Video data sizes continue to grow at roughly 4x per format iteration
as cameras and playback devices are able to capture and display
Deen & Daigle Expires April 27, 2017 [Page 5]
Internet-Draft GGIE Intro October 2016
higher quality images. Early digital video was often captured at
either 320x240 pixel resolution or 640x480 standard definition
resolution. High definition or HD video at 1920x1080 became possible
on some parts of the Internet after 2011, although even in 2016 it
remains unavailable or unreliable through many connections such as
DSL and many mobile networks. Camera and player technologies are
currently expanding again to permit 4K or 3840x2160 pixel resolution
reflecting a 4x data increase over HD.
Streaming is very demanding, requiring consistent frame to frame
playback in consistent constant time. Advanced features such as
pause, fast forward, rewind, slow motion, and fine scrubbing are
considered by users as standard features in players that the network
must support and serve to further the challenge facing the Internet.
New video abilities such as live streaming by users (both one to one
and one to many) bring what has traditionally been done by
professional broadcasters with dedicated broadcast infrastructure
into the realm of every day users with connected smartphones using
the Internet as a real-time global broadcast infrastructure.
5. Historical Approaches to supporting Video on the Internet
5.1. Video as an application
Internet video engineering began by adapting preexisting standards
used for over the air broadcast (OTA) and physical media. Video
encodings, such as AVI and MPEG2, originally designed for playback
from local storage connected to the player where added to the data
types carried by existing protocols like HTTP, and new protocols such
as RTSP and HLS. Early use of the Internet for video was a copy-and-
play model replacing the use of OTA broadcast and physical media to
copy video between systems.
As Internet bandwidth became sufficient to allow delivery of video
data at the same rate it was being decoded, it became possible to
stream video originally at very low resolutions such as 160x120
pixels (19.2 kilopixels), eventually permitting standard definition
(SD) 640x480 pixels (0.3 megapixels), and later high definition of
1920x1080 pixels (2 megapixels). This trend continues with some
providers beginning to offer 4K or 3840x2160 pixels (8.3 megapixels)
requiring very reliable and generous Internet bandwidth end to end
connection between the viewer and source.
Unlike the Web, email, and network file sharing which have been
engineered and standardized in Internet focused organizations such as
the W3C and IETF, video is dependent on standards developed by a very
large number of groups, companies, and organizations which include
Deen & Daigle Expires April 27, 2017 [Page 6]
Internet-Draft GGIE Intro October 2016
the IETF, W3C but also MPEG, SMPTE, CTA, IEEE, ANSI, ISO, networking
and technology companies, many others. In contrast to the extensive
end to end expert knowledge and engineering done to create the Web
and email, Internet video has largely been an evolved cobbling and
adaption exercise done by engineers with their focus on a few, or
one, particular aspect or problem at a time, and little interaction
between other parts of the Internet video ecosystem. While it is
very much possible to deliver video over the Internet, this
uncoordinated cobbling has resulted in many areas of inefficiency
where engineering done from an end to end perspective could provide
the opportunity to vastly improve how video uses the Internet, which
offers the hope of improving the quality of video and increasing the
amount of video which can be delivered.
5.2. Video as a network problem
Network, video, and application engineers have constructed elaborate
solutions for dealing with bandwidth and processing limitations,
network congestion, lossy transport protocols, and the ever growing
size of video data. These solutions commonly fall into one of
several solution types:
1. Reducing data sizes through resolution changes, compression, and
more efficient encodings
2. Downloading before playing instead of real-time streaming
3. Positioning the data close to the viewer via caches, typically on
the network edge
4. Fetching of video data at a rate faster than playback
5. Transport protocols that attempt to deliver video data such that
the data arrives as if it were done on a congestion free/lossless
network
6. Dynamic reselection of sources and transport routes on either a
real-time or frequent intervals, 10-15 seconds, using player
feedback mechanisms or network telemetry
5.3. Video Ecosystem Encapsulation
The current delivery ecosystem for video has been primarily developed
at the higher application layers of the stack. While there has been
some video work done at lower levels such as general-purpose
transport improvements, caching protocols in CDNi, various
multicasting approaches, and other efforts, the majority of video-
specific work has previously been done by groups such as ISO's Moving
Deen & Daigle Expires April 27, 2017 [Page 7]
Internet-Draft GGIE Intro October 2016
Pictures Expert Group (MPEG) which have focused on codecs and codec
transport optimized for use on the Internet. These efforts have made
video possible on the Internet, but they have done so largely while
treating the underlying network as a basic transporter of data. This
has resulted in little information being exposed to the network,
information that could be used to optimize delivery of the video, and
in an architecture that pushes more and more of the intelligence into
an ever more complex and isolated core.
The current video model benefits from a significant amount of
operational, feature, and protocol encapsulation that has come about
due to different groups working independently on the components that
make it up. Like any system in which distinct pieces are well
encapsulated from one another, this means it is possible to engage in
improvements at the networking layer without the need to coordinate
with higher levels of the video architecture.
6. Problem Statement and Solution Criteria
At its most basic the problem to be solved for video delivery is how
to simultaneous maximize all of the following conditions: The number
of viewing devices simultaneously supported by the network; The
quality of video as measured by bit-rate and resolution; The number
of distinct unique streams that can be delivered.
Solution Constraints
1. Bandwidth growth alone is not a solution
2. Codec efficiency improvements alone are not a solution
3. Existing devices, infrastructure, video delivery techniques must
as much as possible continue to be supported and benefit from new
solutions.
7. The Glass to Glass Internet Ecosystem: GGIE
GGIE is an effort to improve video's use of the Internet by examining
the end to end video ecosystem from the glass lens of the camera
through to the glass the screen, and to identify areas of
simplifications, standardization, and reengineering to make better
use of bandwidth enabling smarter network use by video creators,
distributors, and viewers. GGIE is focused on how video uses the
Internet, and not on how it is encoded or compressed. Likewise GGIE
does not deal with content protection. GGIE's scope however does
include creator and viewer privacy, content identification and
recognition as a means to enable smarter network usage, edge caching,
and discoverability.
Deen & Daigle Expires April 27, 2017 [Page 8]
Internet-Draft GGIE Intro October 2016
GGIE benefits from the encapsulation of the video ecosystem elements
enabling it to introduce evolutional features to elements without
disrupting other distinct encapsulated parts.
GGIE is intended to work with a wide variety of video encoding
codecs, and video distribution and transport protocols. While
examples using MPEG-DASH are used due to is pervasive use, GGIE is
not limited to MPEG-DASH or any other video distribution system or
codec.
Beyond improving the simple experience of a viewer using the Internet
to watch linear video, it is hoped that a set of improved Internet
video infrastructure standards will provide a foundation that permits
innovators to create the next generation of Internet video content
(such as multisource personalized composite experiences, interactive
stories, and live personal broadcasting, to name a few).
Due to the very diverse and large deployment of existing video
playback devices and infrastructure, it is viewed as essential that
any evolved ecosystem continues to work with the majority of the
legacy deployment without the need for updates or changes to the
existing ecosystem.
7.1. Related work: W3C GGIE Taskforce
A companion effort ran through 2015 in the W3C Web and TV Interest
Group's GGIE Taskforce. The W3C GGIE group developed a series of
use-cases on discovery, search, delivery, identity, and metadata
which can be found at https://www.w3.org/2011/webtv/wiki/GGIE_TF
8. GGIE work of relevance to the IETF
This section assumes a working familiarity with video creation and
consumption "life cycle". For reference, an overview has been
provided in the Appendix.
8.1. Affected IETF work areas
It is expected that significant improvement is possible in the video
transport ecosystem by modest evolution and adaption of existing
standards for addressing, transporting, and routing of video data
flows between sources and display.
8.2. Example use cases
The following example use case help illustrate the use of the GGIE
core elements
Deen & Daigle Expires April 27, 2017 [Page 9]
Internet-Draft GGIE Intro October 2016
8.2.1. Alternate Source Discovery
Description: A video player is streaming a movie from a CDN cache in
the core of the network. This use case illustrates the use of a
media identifier to query a media address resolution service to
locate additional alternate sources that offer the same movie.
1. The video player user selects a movie to watch from a list using
the player application UI.
2. The video player application has the media identifier of the
movie in the metadata description of the movie. This identifier
is passed to the playback device when the movie selected.
3. The playback device send a search query to the Media Address
Resolution Service (MARS) which includes the media identifier,
and additional query parameters use to filter the results
returned.
4. The MARS server searches its database and returns all the Media
Encoding Networks matching the media identifier and filters the
results using the additional parameters submitted in the query.
Each Media Encoding Network represents a different encoding of
the video.
5. The player then examines the returned list of media encoding
networks and selects, from its perspecitve, the optimal source
for the title.
6. The player then directs its streaming requests to the selected
Media Encoding Network addresses to obtain the video data for the
movie.
7. The video data is decoded and displayed on the screen
8.2.2. Alternate Format Discovery
Description: A video player is streaming a movie, and wants to send
the audio to another device for playback. However, the current video
data being streamed does not contain any audio that matches the
codecs the audio device can play. The audio device uses the core
GGIE services to locate an alternate encoding of the movie that
contains audio it can decode.
1. The user directs the video player to send the audio portion of
the playing video to an external audio device.
Deen & Daigle Expires April 27, 2017 [Page 10]
Internet-Draft GGIE Intro October 2016
2. The video player application passes the media idenfitier for the
video to the audio device as well as the media encoding network
address the video player is using.
3. The audio device begins streaming from the media encoding network
is was given, but discovers the data does not include audio that
is able to decode.
4. The audio device sends a search query to the Media Address
Resolution Service (MARS) which includes the media identifier,
and additional query parameters including the list of audio
codecs and language choice it is able to decode.
5. The MARS server searches its database and returns all the Media
Encoding Networks matching the media identifier and filters the
results to only those matching the language and audio codec
supplied in the search.
6. The audio player examines the returned list of media encoding
networks, selects a media encoding network and begins streaming
data from it.
7. The external audio player decodes the returned movie data and
plays it for the user.
8.3. Core GGIE elements
GGIE proposes three initial fundamental pieces:
1. Media Identifiers which identify the video at the title, or work
level;
2. Media Encoded Networks which are subnets used to reference the
encoded video data;
3. Media Address Resolution Service which maps Media Identifiers for
a title to the Media Encoded Networks containing the encoded
video versions of the title.
These three foundational elements help by exposing information that
can be used in selection in a way that is independent of the video
encoding and video data storage choice. It also enables more
sophisticated video use cases beyond the basic single device playing
a video stream from an origin server over a flow controlled protocol.
Deen & Daigle Expires April 27, 2017 [Page 11]
Internet-Draft GGIE Intro October 2016
8.3.1. Media Identifiers
A Media Identifier is a URI that carries a content identifier system
declaration, and a content identifier from the system that refers
unambiguously to a work, or title. This can be any content
identification system, GGIE does not specify the system used.
For example, a media identifier for a title identified by an EIDR
value would include a declaration that the identifier is from EIDR,
and would additionally contain the EIDR value.
At the application level, such as UI program guide applications,
search engines, and metadata databases, it is the identification of
the work or identity of the video that is typically of interest and
not the encoding, bit-rate, or the location of CDN caches etc. For
example, a UI would indicate that "the Minions movie" as opposed to
"a 15 megabit per second, HEVC encode with high dynamic range and
Dolby encoded 7.1 English audio of the Minions movie". Those
additional technical details are important when choosing a particular
encoded manifestation of the movie for delivery, decode, and
playback, but they are not generally needed as information to be
presented to the user or used to make viewing choices. Such
technical information is used after the user has chosen the title to
watch, but is used by the playback device not the user in selecting
the video. Media Identifiers in GGIE contain only title information,
and not encoding information.
There are many media identifiers in use for both personal and
professional content, with new ones being introduced seemingly
weekly. To try to create a single identifier to either harmonize or
replace the others, repeatedly been proven in practice to be an
impossible task. Recognizing this, the GGIE instead proposes to
standardize a URI which would contain at least two fields: 1) A
scheme identifier; 2) An unambiguous title identifier (note: this is
unambiguous only within domain of the identified scheme).
For professional content, titles are increasingly identified with a
scheme called EIDR that can identify both master versions of works,
and edit level versions. Likewise, advertisments use a scheme called
AD-ID.
8.3.2. Media Address Resolution Service (MARS)
The media address resolution service (MARS) provides bidirectional
mapping of Media Identifiers to Media Encoding Networks. It is
queryable using a query protocol which returns any results matching
the terms of the query parameters.
Deen & Daigle Expires April 27, 2017 [Page 12]
Internet-Draft GGIE Intro October 2016
A Media Identifier alone isn't sufficient to connect a device to a
video data source. The media identifier distinguishes the work, but
not the technical details of an instance of the work such as codec,
bit-rate, resolution, high dynamic range video, audio encoding, nor
does it include information about available streaming sources etc.
The Media Address Resolution Service (MARS) provides this
association. It can be queried with the Media Identifier, and
optional filtering parameters, and will return Media Encoding Network
addresses for instances of matching encodings of the work.
This translation is used commonly in video streaming services today.
The link provided in the program guide UI will include a unique
identifier for the work which is then mapped by the streaming service
backend into a URI containing a network identifier and other info
which point to a caching server and the media data files in the
cache. MARS generalizes this and make it available via query over
the network.
8.3.3. Media Encoding Networks (MEN)
Media Encoding Encoding Networks are arrangements of encoded video
data that are assigned addresses under a shared prefix and subnet
following a scheme appropriate for the encoding used by the video
data. Each Media Encoding Network instance represents a distinct
instance of a set of associated encodings for a work. Different
Media Encoded Network address assignment schemes would be defined
under GGIE to handle different encode data such as MPEG-DASH and HLS.
For example, a single MEN instance would hold each of the different
variable bit-rate encodes for a single encoding of a video If a new
encoding instance of the video was prepared, it would have separate
and distinct MEN assigned to it.
8.3.3.1. Example: Using Media Encoding Networks with MPEG-DASH
A very basic form a video delivery uses persistent connection from a
player to a video file source which then streams the video by
transmitting the video file data, byte by byte in sequence, from the
first byte of the file until the last. This trivial approach
requires the device to know the server IP address and port number to
connect to. Essentially this involved simply transporting the file
from the source to the playback device in byte order.
In practice simple file streaming is not used beyond local device to
device playing in home networks as it doesn't permit dynamic bit rate
selection, source or session fail over, or trick play (pause, skip
forward, skip backward) etc. Instead manifest files contain lists of
available servers holding MPEG-DASH encodings of the larger video
Deen & Daigle Expires April 27, 2017 [Page 13]
Internet-Draft GGIE Intro October 2016
file divided into fragments containing short portions (e.g. 2-15
seconds) of the video called chunks by MPEG-DASH. (GGIE generalizes
the MPEG-DASH chunk term into the more general shards). Each shard
is a distinct file typically named to reflects the video encode it
belongs to, and it's sequence position.
For example the shards for MY-VIDEO might be names MY-VIDEO-001, MY-
VIDEO-002, ... MY-VIDEO-nnn. The player then requests the shards in
the order it wants them over a data transport protocol such as http,
with the translation of the actual data sent in response to requests
for the named shards being handled by the data server.
So under MPEG-DASH the player is sent a manifest file containing the
address of the data server and the shard name to request. The player
then iterates over the available shards in the order desired by the
user. The manifest then contains URI's with the SERVER-ADDRESS and
the CHUNK name. This file can be sent once per video play, or more
commonly is sent at an interval of ~15 seconds to permit the sending
CDN to customize for each player, and to respond quickly to changes
in the network delivery performance and availability.
Each shard request by the device involves a network level server IP
address and port number, and an application level shard name. The
network is thus able to manage the routing of request to the server,
and the routing of the response, but it lacks the information needed
to do anything else to help optimize the video data transport.
GGIE proposes using Media Encoding Networks an evolution of this that
has the benefit of being backward compatible with manifest files,
while enabling the transport network and video ecosystem to have more
information to the network about the video transport flowing over it.
Using Media Encoding Networks for MPEG-DASH will be described in
another Internet-Draft, but the basic proposal is to assign the
shards into a sequence of IP addresses organized to reflect the same
ordering association that the chunk names followed in the MPEG-DASH
scheme. These shard addresses form a Media Encoding Network, and
they expose to the network layer knowledge of the specific video data
being transported between requesting device and the file server
holding the data.
This in practice means that Media Encoding Network addresses refer to
the shard and not the server holding the shard. This then permits
the network to be involved in the routing of the request for the
shard, as opposed to the CDN preparing the manifest file. Among
other benefits, this permits the network to provide path failover
functionality beyond the CDN manifest.
Deen & Daigle Expires April 27, 2017 [Page 14]
Internet-Draft GGIE Intro October 2016
This enables the network to be involved in shard source selection.
Consider the use case wherein the network becomes aware of a local
cache that holds the requested shard, and is closer to the device
than another cache deeper in the network. The network can direct the
request to the local cache and save the transit cost and bandwidth of
sending the request and response exchange with the deeper cache.
This can reduce network congestion as well as deliver faster
transport for the shard to the playback device.
8.3.4. Media Encoding Network Gateways
In this new approach, the server providing the shard data is possibly
better viewed as acting as a gateway to the shard addresses versus
being just a file server. In practical terms, existing CDN caches
can perform this role by mapping the requested shard address to the
on disk file containing the shard. However, new CDN caches can be
developed work directly with the Media Encoding Network scheme, and
can act as smart caches proactively provisioning data within the
Media Encoding Network address space.
9. Conclusion and Next Steps
GGIE seeks to help address this problem by establish standards based
foundational building blocks that innovators can build upon creating
smarter delivery and transport architectures instead of relying on
raw bandwidth growth to satisfy video's growth.
Next steps will include describing the working prototypes of the GGIE
core elements and more exended use cases addressed by GGIE many of
which were defined in the W3C GGIE Taskforce.
10. Acknowledgements
Contributions to this document came from Bill Rose, Gaurav Naik, John
Brzozowski.
11. IANA Considerations
None (yet).
12. Security Considerations
12.1. Privacy Concerns
The assignment of persistent IPv6 Prefixes to MEN permits the video
being streamed to be identified at the network level by observing the
destrination addreses sent from the player to the media gateway. In
situations where it is desired by the user to prevent this level of
Deen & Daigle Expires April 27, 2017 [Page 15]
Internet-Draft GGIE Intro October 2016
observation is necessary to obscure the true MEN prefix of the video
being streamed.
12.1.1. Privacy via VPN
One remediation is the use of a VPN that will encapsulate and hide
the traffic between the player and the streaming cache, or at least
between the trusted network the player resides on and the streaming
cache network. This will make identification of the actual video
title from the open Internet during transit.
12.1.2. Session Prefix Renumbering
Another technique is to have the player and streaming cache remap the
IPv6 prefix for the streaming session to a new prefix. Under such a
renumbering the cache will advertise to the routing layer and respond
to requests sent from the player to the session prefix just as it
would to the original video MEN prefix.
13. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<http://www.rfc-editor.org/info/rfc2119>.
Appendix A. Overview of the details of the video lifecycle
This section outlines the details of the video lifecycle -- from
creation to consumption -- including the key handholds for building
applications and services around this complex data. The section also
provides more detail about the scope and requirements of video (scale
of data, real-time requirements).
Note: this document only deals with streaming video as used by
movies, TV shows, news broadcasts, sports events, music concert
broadcasts, product videos, personal videos, etc. It does not deal
with video conferencing or WebRTC style video transport.
A.1. Media Lifecycle
The complex workflow of creating media and consuming it is
decomposable into a series of distinct common phases.
Deen & Daigle Expires April 27, 2017 [Page 16]
Internet-Draft GGIE Intro October 2016
A.1.1. Capture
The capture phase involves the original recording of the elements
which will be edited together to make the final work. Captured media
elements can be static images, images with audio, audio only, video
only, or video with audio. In sophisticated capture scenarios more
than one device maybe simultaneously recording.
A.1.1.1. Capture Metadata
The creation of metadata for the element, and for the final video
begins at capture. Typical basic capture metadata includes Camera
ID, exposure, encoder, capture time, and capture format. Some
systems record GPS location data, assigned asset ids, assigned camera
name, camera spatial location and orientation.
A.1.2. Store
The storage phase involves the transport and storage of captured
elements data. During the capture phase, an element is typically
captured into memory in the capture device and is then stored onto
persistent storage such as disc, SD or memory card. Storage can
involve network transport from the recording device to an external
storage system using either storage over IP protocols such as iSCSI,
a data transport such as FTP, or encapsulated data transport over a
protocol such as HTTP.
Storage systems can range from basic disk block storage, to
sophisticated media asset libraries
A.1.2.1. Storage Metadata
Storage systems add to the metadata associated with media elements.
For basic block storage, a file name and file size is typical, as are
a hierarchical grouping, creation date, and last-access date. For
library system an identifier unique to the library is typical, as
well as grouping by one or more attributes, a time stamp recording
the addition to the library and a last access time.
A.1.3. Edit
Editing is the phase where one or more elements are combined and
modified to create the final video work. In the case of live
streaming, the edit phase maybe bypassed.
Deen & Daigle Expires April 27, 2017 [Page 17]
Internet-Draft GGIE Intro October 2016
A.1.4. Package
Packaging is the phase in which the work is encoded in one or more
video and audio codecs. These maybe produce multiple data files, or
they may be combined into a single file container. Typically,
creation or registration of a unique work identifier, for example an
Entertainment Identifier from EIDR, is assigned in the packaging
phase.
A.1.4.1. Package Metadata
A.1.5. Distribute
The distribute phase is publishing or sharing the packaged work to
viewers. Often it involved uploading to a site such as YouTube, or
Facebook for social media, or sending the packaged media to streaming
sites such as Hulu.
It is common for the distribution site to repackage the video often
transcoding it to codecs and bitrates chosen by the distributor as
more efficient for their needs. Distribution of content expected to
be widely viewed often includes prepositioning of the content on a
CDN (Content Distribution Network).
Distribution involves delivery of the video data to the viewer.
A.1.5.1. Distribution Metadata
Distribution often adds or changes considerable amounts of metadata.
The distributor typically assigns a Content Identifier to the work,
that is unique to the distributor and their content management system
(CMS). Additional actions by the distributor such as repacking and
transcoding to new codecs or bitrates can require significant changes
to the media metadata.
A secondary use of distribution metadata is enabling easy discovery
of the content either through a library catalog, EPG (electronic
program guide), or search engine. This phase often includes
significant new metadata generation involving tagging the work by
genre (sci-fi, drama, comedy), sub-genre (space opera, horror,
fantasy), actors, director, release date, similar works, rating level
(PG, PG-13), language level, etc.
A.1.6. Discovery
The discovery phase is the precursor to viewing the work. It is
where the viewer locates the work either through a library catalog, a
Deen & Daigle Expires April 27, 2017 [Page 18]
Internet-Draft GGIE Intro October 2016
playlist, an EPG, or a search. The discover phase connects
interested viewers with distribution sources.
A.1.6.1. Discovery Metadata
It is typical for discovery systems to parse media metadata to use
the information as part of the discovery process. Discovery systems
may parse the content to extract imagery and audio as additional new
metadata for the work to ease the viewers navigation of the discovery
process perhaps as UI elements. The system may import new externally
generated metadata about the work and associate it in its search
system, such as viewer reviews, metadata cross reference indices.
A.1.7. Viewing
The viewing phase encompasses the consumption of the work from the
distributor. For Internet delivered video it is typical for delivery
to involve a CDN to perform the actual delivery.
A.2. Video is not like other Internet data
Video is distinctly different from other Internet data. There are
many characteristics that contribute to video's unique Internet
needs. The most significant characteristics are:
1. large size of video data (Gigabytes per hour of video)
2. high bandwith demands (Mbps to Gbps)
3. low latency demands of streamed video
4. responsiveness to trick play requests by the user (stop, fast
forward, fast reverse, jump ahead, jump back)
5. multiplicity of formats and encodings/bit rates that are
acceptable substitutes for one another
A.2.1. Data Sizes
Simply put compared to all other common Internet data sizes, video is
huge. A still image often ranges from 100KB to 10MB. A video file
can commonly range from 100MB to 50GB. Encoding and compression
options permit streaming videos using bandwidth ranging from 700Kbps
for extremely compressed SD video, to 1.5-3.0 Mbps for SD video, to
2.5-6.0 Mbps for HD video, and 11-30Mbps for 4K video.
Still images have 4 dimensional properties that affect their data
size:
Deen & Daigle Expires April 27, 2017 [Page 19]
Internet-Draft GGIE Intro October 2016
1. number of horizontal X pixels
2. number of vertical Y pixels
3. bytes per pixel
4. compression factor for the image encoding.
Video adds to this:
1. frames per second playback rate
2. visual continuity between frames (meaning users notice when
frames are skipped or played out of order)
3. discontiguous jumps between frames such as skipping forward or
backwards to inserting frames from other sources between
contiguous frames (advertisement placement)
Each video format roughly increases by x4 the data needs of the
previously resolution: (1) SD is 640x480 pixels; (2) HD is 1920x1080
pixels; (3) 4K is 3840x2160 pixels.
Video, like still images, assigns a number of pixels to store color
and luminance information. This currently evolving alongside
resolutions after being stagnant for many years. The introduction of
high dynamic range videos or HDR has changed the color gamut for
video and increased the number of bits needed to carry luminance from
8 to 10 and in some formats more.
Compression is often misunderstood by viewers. Compression does not
change the video resolution, SD is still 640x480 pixels, HD is still
1980x1080 pixels. What changes is the quality of the detail in each
frame, and between frames.
Video is in its simplest form a series of still images shown
sequentially over time, adding an additional attribute to manage.
A.2.2. Low Latency Transport
Viewers demand that video plays back without any stutter, skips, or
pauses, which translates into low latency, high reliability transport
of the video data.
Deen & Daigle Expires April 27, 2017 [Page 20]
Internet-Draft GGIE Intro October 2016
A.2.3. Multiplicity of Acceptable Formats
One of the unique aspects of video viewing is that there can exist
multiple different encodings/versions of the same video, many of
which are acceptable substitutes for one another. This is a unique
aspect of video viewing and differentiates video delivery from other
data transports.
Other application data types don't have or leverage the concept of
semantic equivalences to the same extent as video. Even email, which
supports multiple encodings in a multipart MIME message, has a finite
number of representations of "the message", shipped as one unit,
whereas video often has many distinct different encodings each as
separate file or container of files managed as a distinct entity from
the others.
A.3. Video Transport
A.3.1. File vs Stream
There are two common ways of transporting video on the Internet: 1)
File based; 2) Streaming. File based transport can use any file
transport protocol with FTP and BitTorrent being two popular choices.
File based playback involves copying a file and then playing it.
There are schemes which permit playing portions of the file while it
progressively is copied, but these schemes involve moving the file
from A->B then playing on B. FTP and BitTorrent are examples of file
copy protocols.
Streaming playback is most similar to a traditional Cable or OTA
viewing of a video. The video is delivered from the streaming
service to the playback device in real time enabling the playback
device to receive, decode, and display the video data in real time.
Communication between the player and the source enable pausing, fast
forward and rewind by managing the data blocks which are sent to the
player device.
Authors' Addresses
Glenn Deen
NBCUniversal
Email: rgd.ietf@gmail.com
Deen & Daigle Expires April 27, 2017 [Page 21]
Internet-Draft GGIE Intro October 2016
Leslie Daigle
Thinking Cat Enterprises LLC
Email: ldaigle@thinkingcat.com
Deen & Daigle Expires April 27, 2017 [Page 22]