Audio/Video Transport Working Group Minutes
Reported by Magnus Westerlund and Colin Perkins
The AVT working group met once at the 57th IETF in Vienna. During this
session the working group discussed the updated charter, current status,
VoIP congestion control concerns raised by the IAB, the SSM RTCP
Extension and numerous RTP payload formats.
Introduction and document status
The meeting started by Stephen Casner announcing that this will be his last
meeting as AVT co-chair. Allison Mankin presented Stephen with a plaque
enumerating the RFCs that Stephen has been shepherding during his time as
chair for AVT since its start in 1992. The working group showed its
appreciation for his long service with thundering applause.
Stephen Casner outlined the current working group status. No RFCs
relating to AVT have been published since the last meeting, however there
are a currently 7 in the RFC editor's queue. The RTP specification is
currently in author's 48 hours and has gone through one substantial
change. The usage of the SDP RR and RS bandwidth modifier results in
change to the RTCP interval calculation mechanism. An RTP active sender is
allowed to take its part of the RR part when the number of senders are
larger than RS/(RS+RR). This is a generalisation of the previous 25% rule
that was present. The senders=0 clause in the example algorithm (A.7) has
also been removed to prevent senders bandwidth bleeding to receivers when
RR=0.
The other AVT drafts in the RFC editor's queue are the AVP profile, the RTP
profile MIME registration, the SDP bandwidth modifier for RTCP, the RTP
payload format for EVRC/SMV and the RTP payload format for ETSI ES 201 108
DSR.
The specification for Enhanced IP/UDP/RTP header compression is with the RFC
Editor, but the authors would like to fix one technical shortcoming
before publication. This can be done if the Area Director and Working
Group approve of the change. The issue is to include the CSRC list in the
calculation of the Headers Checksum (HDRCKSUM), to enable header
validation and prevent error propagation in the CSRC list. The working
group was given until 18th July (the end of the IETF week) to comment on
this change. Positive comments were made during the meeting, and there were
no objections on the mailing list.
The working group also has a number of drafts in IESG review:
Tunnelling multiplexed compressed RTP (TCRTP), Secure RTP (which has had
some recent modifications in response to AD review), the Payload format for
MPEG-4, RTCP extended reports, RTCP feedback
(draft-ietf-avt-rtcp-feedback-07) and associated simulations
(draft-burmeister-avt-rtcp-feedback-sim-02), and RTP
retransmission
(draft-ietf-avt-rtp-retransmission-08).
The drafts of Uneven Level Protection
(draft-ietf-avt-ulp-07.txt) and Unequal erasure protection
(draft-ietf-avt-uxp-05.txt) were not updated for this meeting.
Colin Perkins noted that ITU-T SG16 Q.14 is working on standards for FAX
over RTP. He expressed interest in increasing coordination between ITU-T and
AVT to ensure that solutions developed do not conflict. He also noted that a
MIME type registration must be done to ensure that the FAX over RTP work can
be referenced by IETF protocols - such a registration should be
coordinated with AVT.
Charter Update
A preliminary updated WG charter with milestones was presented and has also
been posted to the mailing list. Please review and send any comments to the
list. The updated charter contains the following items:
- Advance RTP and the A/V profile to Full standard RFC - Review and
revise RTP payload formats for Draft Standard RFC, as appropriate -
Develop new RTP payload formats as needed. - Complete the generic FEC
formats (ULP & UXP) - Complete the RTCP extensions for SSM sessions -
Develop a framing method for use of RTP/RTCP over TCP & TLS - Review
applicability of RTP header compression for MPLS - Develop a new RTP
profile combining AVPF and SAVP - Coordinate with the DCCP working group to
ensure efficient transport of RTP over DCCP, and to ensure an
appropriate API is developed
Longer term goals of the working group are to move to Draft Standard the
SRTP Profile, the Extended RTP Profile for RTCP-based Feedback, the
Compressed RTP (CRTP) framework, including Enhanced CRTP, and the RTP MIB.
Further development of media codecs is not expected to be chartered.
DCCP
http://www.ietf.org/html.charters/dccp-charter.html
Aaron Falk made a short informational speech about DCCP and its
upcoming design review. He encouraged people to read the
specification, and come to the design review. He especially desired
people in AVT to comment on any API related issues, as this relates to the
possible usage of DCCP.
RTP Payload Format for MIDI
draft-ietf-avt-mwpp-midi-rtp-08.txt
draft-lazzaro-avt-mwpp-coding-guidelines-03.txt
Colin Perkins outlined some slides from John Lazarro regarding the RTP
payload format for MIDI. This format has been reviewed by the MMA (the MIDI
Manufacturers Association) and their comments have been addressed. An
implementation exists and will be updated in regards to the changes
during the summer. It is expected that the draft will be ready for
working group last call before the next meeting.
The companion implementation guide for RTP MIDI has been updated to more
concern the MIDI parts, like recovery journal, than to deal with general RTP
things. John Lazzaro has requested that this document becomes a AVT
working group document. A hum was performed with some indication for
approval and none against.
RTP framing for TCP and TLS
draft-lazzaro-avt-rtp-framing-contrans-01.txt
Colin Perkins updated the group on the RTP framing for TCP and TLS. This
draft has been awaiting resolution of the MMUSIC WG comedia work and the
rechartering. It can be expected to make progress once the new charter is
accepted.
IAB Concerns Regarding VoIP congestion control
draft-iab-congestion-00.txt
Sally Floyd made a short presentation on an initial draft of IAB
recommendations regarding congestion control for Voice over IP flows. The
basic recommendation is that any flow that is facing persistent high rates of
packet loss should stop sending. Having congestion control of some type
will ensure that congestion collapse does not happen, user quality can be
maintained, and that fairness against other VoIP flows and TCP flows can be
maintained in a best effort network.
The draft does not recommend any particular deployment path for VoIP, like
best effort, QoS reservations, etc. It also not specifying any
behaviour, it is only a recommendation that there is a need to address
these issues. It also tries to address some of the framing problems and
what assumptions that can be made, like what is fairness, and is the
network, bit-rate or processor limited.
Several questions and comments where raised. One was the need to clarify
what it means to terminate a connection: in this case stop sending media
(not necessarily to terminate any signalling relationship). It was asked if
RTCP information is sufficient for deciding when to terminate the data
flow, which will need further investigation.
Is was also asked why the draft is restricted to VoIP and doesn't
address other media types? This is not because other media don't need
congestion control for other media, rather it's a reaction to the
special treatment often proposed for voice (e.g. because flows are low
bandwidth). This special treatment is not considered appropriate. It was
also clarified that the document does not address congestion control for
multicast.
RTP Payload for DTMF Digits, Telephony Tones and Telephony Signals
draft-ietf-avt-rfc2833bis-03.txt
Henning Schulzrinne presented the current status of the draft. No major
technical changes have occurred, and it is now believed that all events
have been tracked down (however, review from someone working in this area
would be appreciated). The examples in the draft have also been
improved. Henning believes that the document is ready for WG last call,
however some more dedicated review would be beneficial. Tom Taylor and
Flemming Andreasen volunteered to perform a review of the document. The
goal is to have the draft ready for last call before or around the next
IETF meeting.
This draft was originally intended to become a Draft Standard RFC,
however so far the interop testing has not progressed. Henning
questioned the level of interoperability testing that will be needed: if
every event needs to be tested that will be a futile effort as they are
over a 100 in total, and from several different systems.
Alternatives are:
- interop testing only needs to show that the basic timing and
redundancy mechanisms works - to strip all events that can't be
tested, which probably will result in only DTMF tones will be left - the
last alternative is to leave this as a proposed standard for all future.
The chairs will ask the area directors if basic timing and redundancy
interoperability is sufficient for testing for draft standard. It was
proposed that the draft be continued to be progressed to WG last call and be
published as Proposed Standard RFC, while work continues on
interoperability testing. The only consideration is if any of the
interoperability will result in technical changes to the document, in that
case it might be better to wait. It would be advantageous if someone
actually doing implementation work using this format could be
participate in the interoperability testing.
Payload format for Speex Codec
draft-herlein-speex-rtp-profile-01.txt
Phil Kerr presented an update to the RTP payload format for Speex. The
format has received extensive comments on the mailing list that need to be
addressed in next version. Comments in the meeting were regarding the
usage of the RTP timestamp rate. There has been some consideration that a
codec parameter for rate may be redundant, since the codec has some
functionality for recovering sampling rate. Another alternative would be to
use different RTP payload types and timestamp rates depending on codec
rate, this is however not without problem as Stephen Casner pointed out. The
issue needs further investigation. The recovery of the timestamp for
different frames when performing aggregation also needs to be
clarified.
Payload format for Vorbis
draft-kerr-avt-vorbis-rtp-02.txt
Phil Kerr presented an update, where the main issue is related to the
codebooks that the codec uses. One thing introduced in the last version is
caching of the codebooks. This needs to be clarified, especially in cases
where the code book needs to transfered. A codebook is not limited in size
but they are commonly in the size 10-14 kbytes. So far suggested
mechanism are: 1) Transfer in HTTP, 2) send it RTCP APP packets, 3) Use RTP
over TCP, 4) try to define a limited set of codebooks that has to be
cached and are simply referenced. Steve Casner commented that in RTP
version 1 there was a mechanism to transfer parameters in parallel.
However this was removed due to synchronisation problems. Having any such
mechanism is a slippery slope and needs to be threaded carefully.
What is the gain from performing application level fragmentation, rather
than using IP level fragmentation? Need to be further clarified and
evaluated if enough gain is present to warrant the complexity of the
mechanism. There has also been a number of other general comments on the
draft on the mailing list regarding usage of SDP parameters, Path MTU
discovery, RTP timestamp recovery, and frame aggregation that needs to be
resolved.
Internet Low Bit Rate Codec and RTP payload format
draft-ietf-avt-ilbc-codec-02.txt
draft-ietf-avt-rtp-ilbc-02.txt
Alan Duric presented an update of the codec and payload format. The codec
has been reviewed by speech coding experts and their comments will be
distributed when the last call is announced. As a result of the review the
codec specification has been clarified and corrected. The new version
should be possible to implement based only on the description rather then
the source code.
The payload format has an open issue with mode flag if only one side is
capable of supporting 20 ms frame length in an SDP Offer/Answer
negotiation. The proposal to fix this is to include an offer/answer
section in the draft, specifying bi-directional behaviour for the mode
flag. A related issue is that ptime can't double as mode flag (this
should be explained in the text, since it is a common question).
As there has been no demand for comfort noise (CN) generation it will not be
included in the Codec. Henry Sinnreich commented that Voice Activity
Detection (VAD) and CN generation is important. The counter comment was
that an external VAD and the generic CN can be used anyway. There might
also be future work to provide iLBC with VAD and CN.
Colin Perkins asked if there is any need for test vectors? The answer from
Alan is that the codec specification is desired to remain being a
floating-point defined one and therefore exact test vectors can't be used.
However there is a possibility to provide test vector and then use PESQ
measurements to check how far from the nominal output ones
implementation produces. If any vector will be provided they will be in a
separate draft.
Jon Peterson asked if the mode parameter is a more generic parameter and
therefore should be defined as such. Alan said that he had seen it in some
other codecs. Magnus Westerlund commented that any mode parameter and
their values are codec specific and should therefore be defined as such.
RTP Payload Format for Uncompressed Video
draft-ietf-avt-uncomp-video-03.txt
The current status and remaining open issues were presented by Ladan
Gharai. There have been few technical changes since the last version,
mostly editorial and related to understanding. The two payload
parameters sub-sampling and color-mode has been combined into one
parameter (sampling) to avoid the usage of "impossible"
combinations.
The remaining open issue is if the currently defined colorimetry values are
sufficient or if there are more that should be defined? David Singer also
had an comment that YUV-4:4:2 sampling can have their chroma samples in
different orders. This needs to be addressed.
After the next update the draft is expected to be mature enough for
working group last call.
RTP Payload for H.264/JVT Video
draft-ietf-avt-rtp-h264-02.txt
Stephan Wenger presented the current status. The video codec itself has now
been approved by ITU as H.264 and will soon also be approved by
ISO/IEC. Work will continue in the JVT for professional extensions,
however the NALU interface will not be changed, so this work is not
believed to effect the payload format. The updated draft has been
significantly improved regarding understanding, especially of the
decoding order number (DON) concept.
There are a few remaining open issues. Should we do base64 encoding of the
parameters sets instead of base16? There does not seem to be any formal
problem of using this, and it will provide better efficiency. The
security consideration needs to be improved and carefully reviewed. There is
concern over Timing SEI messages, since broken implementations may
disregard the RTP timestamp. There is need for strong
clarifications on the problem of doing that. Steve Casner commented that at
least the blame and cause are on the same side, namely with the
decoder.
RTP Payload format for 3GPP Timed Text
draft-rey-avt-3gpp-timed-text-00.txt
Jose Rey presented for the first time a proposal for an RTP payload
format for the 3GPP timed text format. The timed text format provides
decorated and timed text, that can be used for karaoke and
subtitling. There is a need to complete this work fairly quickly as the
3GPP Release 6 deadline is officially march 2004, and may move to June
2004.
Steve Casner commented that the current proposal seems to provide much
flexibility that does not provide any useful functionality, and only
complicates the format. Colin Perkins also questioned how this work
relates to RFC 2793, RTP Payload for Text Conversation. The answer is that
text conversation does not provide modifiers for style and fonts and other
decorations as 3GPP Timed Text provides.
Joerg Ott raised the most substantial question: should this be a more
generic work. The applications are not limited to 3GPP mobile phones with
streaming, and a generic format seams appropriate. David Singer
commented that he had hoped that the W3C and MPEG work on authoring and
transport of decorated and timed text would be further progressed at this
time, and could provide input into this work. However Dave also pointed out
that there is a strong need to meet the 3GPP deadline. Magnus
Westerlund further commented that he hoped that people were talking about a
generic payload format rather than specifying a generic set of
operations on text (i.e. a text codec) in duplication of work ongoing in
other standards bodies. David singer further commented that there are
usually no comments when people standardise multiple different codecs, that
one generic is enough. The conclusion is that there is need to further
explore the scope of the work here. The subject should be clarified
through a mail discussion or internet draft.
The request for working group status has been delayed until the open
issues regarding how generic the work should be can be resolved.
RTCP Extensions for SSM Sessions with Unicast Feedback
draft-ietf-avt-rtcpssm-04.txt
Joerg Ott presented the update of the draft. The format has reverted back to
using its own RSI RTCP format instead of defining XR extension formats.
Some editorial issues have not been fixed yet. The authors believe there are
only a few remaining technical issues for this draft.
One issue is if the distribution source should be allowed to use only the
senders RTCP bandwidth part or also the receivers? Steve commented that the
appropriate thing would be to allow the distribution source to use all RTCP
bandwidth as it summaries all RTCP. Magnus Westerlund also commented that
care should be take to define the RTCP behaviour for SSM in the draft as the
default, however future profiles may redefine this behaviour.
The second issue is the possible need to have values indicating for which
sources, or number of sources, the gathered information in RSI mode
applies to. This may be an unnecessary complication of the format, and
Magnus Westerlund commented that it only seems to be applicable to some of
the formats. Joerg asked if there is any need for this feature
otherwise the authors plan to go for simplicity here (meaning that the
receivers have to trust the sender to do the reasonable thing when
summarising).
The third issue is that RTCP BYE packets are reported five times (as is
done for SSRC collisions). However there seems to be little necessity of
doing this for normal RTCP BYE packets. Steve Casner commented that it
seems to be an advantage of having this to work in the same way as
standard RTCP. Colin commented that there is probably no need to report a
collision more than one every time you detect it: if the receiver that
causes the collision does not receive the report packet, the collision will
continue and a new collision report will be sent after the next
observation of it by the source.
There is significant interest in this work and the authors should do their
best to meet the December milestone in the preliminary revised charter.
Colin Perkins pointed out that the definition of the "a=rtcp"
attribute contradicts the definition in
draft-ietf-mmusic-sdp-nat-05.txt. The intention from the authors was to use
it as defined in the SDP NAT draft.
Colin Perkins noted that while the Security Considerations section has a
thorough declaration of the problems present with this RTCP
extension, there is a need to have normative declaration on how these
security considerations shall be addressed in different
applications. If not, there will not be possible to have an
interoperable implementations.
Requirements for performing RTP header compression over MPLS networks
draft-ash-e2e-voip-hdr-comp-rqmts-00.txt
draft-ash-e2e-vompls-hdr-compress-01.txt
draft-ash-e2e-crtp-hdr-compress-01.txt
Jerry Ash presented an introduction and motivation for the work that has
previously been discussed in the MPLS working group. There are several
ideas that has been proposed in this context. The main idea is to
provide RTP/UDP/IP header compression over several hops in access
networks on in backbones that have limited bandwidth or expensive
bit-rate costs. Ideas include that one can perform header
compression from an ingress point in an MPLS network to the egress point
without doing compression/decompression for each link. Instead the MPLS
router can forward the compressed packets based on either MPLS labels, or
the compression context IDs. However the scalability of each solution
needs to be further investigated. The context switch idea could also be
applied outside an MPLS network for other links. Another part of the
complete framework is to ensure that the header compression mechanism can
handle the long delay that may result between the compression and
decompression points. So far the assumption is that ECRTP will be enough for
this.
There were several voices raised questioning the need of this work, but
also some comments in support. There seems to be interest to migrate
legacy voice channels done over circuit switched connections to instead use
VoIP, however there was confusion in regards to what "End-to-End" in the
presentation meant. However this is either MPLS ingress to egress, or in
some optimised cases may be through end-to-end if all nodes from
end-point to end-point support SCID switching and header
compression. This mechanism will not be media dependent as it will be able to
compress the header for any RTP transported media. Although an initial
list of tasks was presented, some people expressed concern that the scope of
the work is not clear and that the rationale is poorly defined. It is
apparent that the work must start by writing a requirements document which
defines the goals of the work, and reach agreement on that.
Co-coordination with other working groups will be needed, in
particular the MPLS working group, and it is hoped that the people from
other WG will participate in also in the initial phases defining
requirements.
- * -
|