Minneapolis fecframe session 1 notes
Chairs Ð Marshall Eubanks, Greg Shepherd
Note taker Ð Bill VerSteeg
Agenda bashing -
Magnus WesterlundÐ mmusic needs to be on the conflict list
Chairs Ð thought we had done this, will go back to powers that be and coordinate with A-Ds
Draft-ietf-fecframe-framework Ð Mark Watson
Mark walked through changes from framework-02 to framework-03
FEC schemes specify RTP transport flows
Feedback via RTP/RTCP
Magnus Ð how do we handle feedback for non RTP flows?
Marshall- no desire to reinvent RTCP
Magnus- Also need to consider aggregate flows
Mark Ð walking through presentation - added optional RTP demux for FEC and source, followed by RTP processing of source
Marshall Ð what is status of framework? Once we address MagnusÕ concerns, is this ready for last call? No objections noted
1-D and 2-D Parity FEC Ð Ali Begen
Ali walked through introduction to 1D and 2D FEC, including brief overview of XOR based FEC
No backward compatibility, but full RTP compliance
Magnus Ð is RTCP CNAME adequate for associating flows? Maybe require same SSRC for both flows
Marshall- What would we be multiplexing on?
Magnus Ð port or address , depending if the flows are on the same or different RTP sessions. Need to be careful of SSRC collisions
Marshall Ð need to describe desired SSRC behavior in the FEC documents rather than refer it to AVT
Bill VerSteeg Ð suggest refer CNAME correlation to AVT rather than fixing it in fecframe
Ali and Magnus Ð discussion of mappings done in retransmission flows and how it compares to the CNAME problem under discussion for FECFRAME Ð
Bill Ð we should try to converge unicast and multicast repair flows, as well as FEC and explicit packets. A common CNAME convention between the models is desirable
Ali Ð continuing with header discussions, IANA registrations, SDP, and SMPTE liaison from presentation
Marshall Ð VSF work should be a normative reference to the IETF work Ð
Ali Ð not using the SMPTE 2022-1 spec, not backwards compatible, so a normative reference is not advisable
DVB AL-FEC Ð Ali Begen
Ali walked through the 1-D baseline parity code and Raptor enhancement layer
Baseline mandatory, enhancement optional
Old spec was UDP for Raptor, RTP for parity Ð new spec is RTP for all
Outline of differences between RFC 3550 and SMPTE 2022-1
Outline of resolution plan for aligning IETF and DVB
Outline of IANA registration issues
Marshall Ð request update on DVB liason
Ali Ð New RTP profile by DVB may fix SSRC, CSRC and Ò96Ó issues Ð timestamp issue remains, though Ð fielded systems present problems to changing the DVB spec
Magnus and Ali Ð discussing merits of converging DVB and IETF vs letting DVB move independently Ð suggest completing IETF effort and recommend to DVB to reference the new draft
FEC Grouping Semantics in SDP Ð Ali Begen
Ali walked through presentation - Similar to mmusic presentation
Describe the relationships between the various flows, and the required SDP to describe those relationships.
Describe the new semantics for additive and priority relationships
Describe problems for equal-priority and equal additive properties
Marshall Ð compare to BGP solution for priority
Ali, Greg Ð motivation for priority dynamically specified rather than a-priori assignment
Dave Oran Ð what is motivation for including flow priority in SDP?
Marshall, Ali, Marshall Ð discussion of additive FEC that requires ordering of FECs vs associative ordering Ð no non-systematic codes, so each FEC could be useful in the absence of other flows Ð flow ordering is for efficiency
Mark Watson Ð Can always use a lower priority FEC even though a higher priority flow is lost Ð priority was designed to allow massive use of smaller number of flows
Dave Ð network resources should not be primary driver for this design
Marshall/Dave/Ali Ð maybe we should call this preference rather than priority- but SHOULD/MUST gets wrapped up in Òpriority/preferenceÓ discussion
Mark/Ali/Marshall/Greg Ð is the discussion around FEC priority an SDP issue or a framework issue?
Rajiv Asati Ð ÒpreferenceÓ is better than ÒpriorityÓ Ð equal priority should be deterministic, perhaps via instance ID
Ali Ð need to have this functionality
Config Signaling Ð Rajiv Asati
Draft 00 to draft 01 Ð minor textual updates regarding usage
Suggestion for last call after framework document goes to last call
FECFRAME Ð Raptor Ð Mark Watson
Mark walks through overview document
3 FEC schemes for Raptor
3GPP protect multiple source flows, using packet tagging to identify source packets
DVB optimized Raptor for arbitrary flows Ð restricted set of block sizes
DVB single sequenced flow
Alignment with recent DVB changes
Defines RTP header fields as per DVB
Proposed for adoption as working group item
Ali Ð are we supposed to define the RTP format parameters in the SDP?
Mark Ð 2 options Ð RTP payload format specific SDP or generic format
Magnus Ð is torn between define payload specific format and generic format Ð perhaps a generic framework that allows for specific portions?
Mark Ð isolate FEC SDP parsing from general flow SDP parsing
Magnus Ð differentiate general parsing from FEC parsing Ð take this issue to a cross discussion between AVT and FECFRAME
OPEN DISCUSSION
Greg Ð want to discuss feedback using RTCP
Magnus Ð should use RTCP for feedback Ð need to consider cases that do not use RTP
Marshall- need to consider FLASH and Windows Media, which do not use RTP