Transport Area Report - Orlando FL
Audio/Video Transport (avt)
The main area of discussion in AVT during this meeting was multiplexing of RTP streams. The primary motivation for this is the reduction in header overhead when trunking large numbers of PSTN calls across an IP network, but there is also desire for a multiplexing solution from the MPEG-4 community. It is unclear whether a single protocol can satisfy both aims, and which of the five proposals currently submitted to the group will go forwards. The group decided to conduct more simulation and analysis of the proposals, in order to facilitate a fair comparison.
The revision of RTP towards draft standard status is proceeding. The RTP spec itself is now essentially complete, the A/V profile needs more work. In particular, some progress was made towards deciding how to register codecs in the MIME namespace, and Philipp Hoschka has volunteered to take the lead on this effort.
Considerable progress has been made on the definition of a payload format for MPEG-4 content, in conjuction with the MPEG committee. The preferred solution is a single payload format for MPEG-4 SyncLayer packets (mapped onto RTP), which avoids many issues with DMIF and FlexMux which were previously thought to be problematic. An internet draft is expected in the next few weeks summarising progress here. A "generic" payload format, primarily intended for transport of MPEG-4 elementary streams was also presented, but would depend upon the elementary stream interface being normative.
Payload formats for DTMF tones, interleaved audio, Reed-Solomon coding and parity FEC were presented. The parity FEC draft is ready for last call, as will be the DTMF draft after one more revision. The RTP MIB and guidelines for RTP payload format writers drafts are also ready for last call. A revision of the PureVoice payload format draft was not posted as expected; the status needs to be confirmed with the authors.
The updated charter was presented. This will be submitted to the IESG in the near future.
Differentiated Services (diffserv)
The working group endorsed the EF (expedited forwarding) draft as updated but did not reach final consensus on the AF (assured forwarding) draft where another revision and WG last call will be required. For the time being, other PHB proposals will need to use experimental code points, which do not require WG approval. The framework draft was reviewed and agreed to need further work. The draft on interworking with RSVP was reviewed positively and will be progressed in the IETF, but probably not in the diffserv WG itself. More work is needed on diffserv "edge" devices such as conditioners. A draft on diffserv/MPLS interworking was presented and was agreed not to be an item for the diffserv WG itself. The WG agreed to define a diffserv MIB.
IP Performance Metrics (ippm)
The working group opened by thanking Guy Almes for his good work as Co-Chair, and welcomed Matt Zekauskas as incoming Co-Chair. We began with a status report: connectivity has been approved as an Experimental RFC; 1-way packet loss and delay have had WG Last Call but will be delayed a few weeks to get more input from T1A1 and ITU-T; round-trip delay is nearly ready for WG Last Call; an initial framework draft for bulk transfer metrics has been written but is incomplete; drafts on jitter and other synthetic metrics exist but need work to harmonize with ITU and to achieve more WG consensus.
The first half of the meeting was devoted to hearing a report from Garry Couch, the Chair of T1A1.3, discussing coordination of T1A1 and ITU-T work on IP performance with IPPM's work. Several clear principles emerged, including the basic agreement of our frameworks and some delineation of responsibilities. (For example, ITU-T makes "goodness" criteria, while we focus on methodology and never define "acceptable" results.) We will work with Garry to quickly document the mapping between the ITU and IPPM frameworks and bring the document to both the ITU and IETF.
The second hour was devoted to short reports on the current metrics drafts in progress: 1-way loss and delay (Matt Zekauskas), round-trip delay (Guy Almes), bulk transfer (Matt Mathis), IP delay variation ("jitter") (Carlo Demichelis), and other synthetic metrics derived from the base definitions (Rajeev Koodli). The bulk transfer framework presentation and the derived metrics generated the most comments. With regards to bulk transfer, the framework idea seems like a good one, but there must also be emphasis on getting metrics within the framework out, and have these be metrics that do show real-world problems (we would like to know if buying more capacity might reduce our throughput). With regard to derived metrics, there was caution to ensure they were general (e.g., not just a "DiffServ metric"); perhaps a series of BCP documents would suffice (e.g., applying delay to DiffServ).
IP Telephony (iptel)
The primary focus of the working groups work is the gateway location protocol. There was a discussion of open issues, many of which were resolved. In particular, it was agreed to keep QoS out of the protocol, and provide a dimensionless "capacity" metric to characterize gateways. There was discussion of the general aggregation vs. attribute issue, with no real conclusions. Cost metrics were discussed. There was agreement to not specify complex cost structures, but not any firm agreement on whether to specify type/length/value, a URL, or some other thing, or whether to define some basic structures now. There was consensus to allow either colocation or separation of the signaling server and location server.
There was a presentation on PGRP, the Peer Gateway Routing Protocol, as an intra-domain solution. The discussion was very low level, detailed answers to questions raised on the list. The chair emphasized that the working group is working right now on an inter-domain protocol. There was a presentation by Lev Slutsman proposing to effectively integrate PSTN and IP call routing for end to end optimality. It was agreed that such harmonization of IP and PSTN protocols was unachievable and outside the scope of the working group.
The work then shifted to the call processing language. There was a presentation on open issues in the call processing language, which generated little discussion. The chair has growing concerns about the lack of participation of the group in this work, with really only one organization working on it (Columbia U.).
Integrated Services (intserv)
The intserv working group did not meet in Orlando.
Integrated Services over Specific Link Layers (issll)
The ISSLL working group held one meeting, with blue-sheet attendance of 128. The order of business included:
- A discussion of outstanding problem with receiver heterogeneity in IS802 SBM raised at working group last-call, and proposal for moving forward.
- A brief presentation of ISSLOW service mapping document, and request by authors for WG last call.
- An informational presentation about IEEE 802.11 Wireless LAN QoS capabilities, and possible ISSLL work for this technology.
- Presentations and discussion regarding use of Diffserv clouds as one component of an end-to-end signalled QoS path supporting Intserv services.
- An informational presentation by Khaled Amer, chair of the newly forming IEEE Study Group on QoS and Flow Control, regarding opportunities for collaboration.
- A brief presentation from Michael Smirnov passing on the ION chair's request for comments about a presentation Michael will make in the ION meeting.
Expected working group actions going forward from the meeting:
- Immediate publication of Version 1 IS802 documents (framework, SBM, mapping).
- Development of backward compatible extensions for SBM receiver heterogeneity, based on proposal presented at the meeting.
- Solicit reviewer comments, then issue a working group last call for ISSLOW mapping document.
- Work on Intserv-using-Diffserv framework and service mapping documents - framework will be based largely on currently existing internet drafts.
- Update to the working group charter to reflect completed and current work items.
Multicast-Address Allocation (malloc)
The malloc working group met on Thursday, December 10, 1998. Ramesh Govindan presented a preliminary MASC deployment plan. Steve Hanna described changes required to add language tags to scope names. Munil Shah described changes to MDHCP since IETF 42. These presentations were well received, with few comments. Dave Thaler led a discussion of MDHCP open issues. Several of these were contentious, but all were eventually resolved with rough consensus (which will be confirmed on the mailing list). With this achieved, we plan to revise the MDHCP spec (and rename it) and move it to WG last call in January 1999.
Multiparty Multimedia Session Control (mmusic)
Mark Handley briefly mentioned the ID on the Conferencing Architecture - this draft has been updated, and will be resubmitted to the ID archive shortly after the IETF. We plan to issue WG last call on this for Informational RFC in January.
Mark Handley presented the status of SAP and the SAP security draft. There has been disagreement between the authors on the use of public key encryption algorithms in a mode that effectively turns them into symmetric algorithms. The arguments for and against this were presented in the hope of getting consensus from the group. Very little feedback was forthcoming.
Colin Perkins presented a modified version of SAP that can support IPv6. The most contentious issue is whether SAP should be able to carry payloads other than SDP. This is a tradeoff between flexibility and interoperability. There was some discussion, and a new draft will be forthcoming.
Mark Handley presented the changes to the SIP spec that were requested by the IESG. These were not controversial. The group was asked whether they believe they need a new WG last call to review these changes. No-one wanted a new WG last call.
Jonathan Rosenberg presented an internet draft on SIP provisional response reliability. There was not consensus on this draft, but there did seem to be resonable consensus that the problem is worth looking at.
Jonathan Rosenberg presented an internet draft on SIP CGI. There seemed to be consensus that we did not want to pursue this in MMUSIC.
Paul Christ presented work on RTSP-based stream control for MPEG-4. There was no discussion, perhaps because the presentation was difficult to follow.
Joerg Ott presented an update to the message bus, which had been discussed at the previous IETF. It has not yet been decided whether this should be an MMUSIC work item. There was also significant discussion about whether this should be modified to permit wide-area operation (currently it's purely intended for local coordination). The authors are reluctant to turn it into a general purpose conference control protocol.
Network Address Translators (nat)
"NAT Terminology" and "Traditional NAT" drafts completed WG last call prior to the meeting and await IESG review for advancement to informational RFCs. There were a total of 5 new NAT drafts reviewed in the meeting. This included a draft summarizing security model for NAT domains, a draft on Host-NAT protocol and a draft outling the framework for external agents to interface with NATs. Two other drafts titled "NAT friendly application design guidelines" and "A multihoming solution using NATs" were also reviewed. The ADs have advised the WG to review Host-NAT protocols and issues, while ensuring that the base documents are given priority. The WG is on-track to date with regard to meeting WG milestones. There were positive comments from audience for addressing security issues through Host-NAT, security Model and framework drafts. An agreement was reached to rename "NAT Protocol Issues" draft as "NAT Protocol Complications". There was a proposal to send a request to all WG chairs soliciting input for this draft from the wider IETF community. The draft has been slow to progress due to lack of sufficient input.
Network File System Version 4 (nfsv4)
The two 1-hour sessions were structured with a series of short presentations followed by a period for open discussion. Brent Callaghan presented an overview of the server namespace proposal that lets clients browse the server with LOOKUP and READ requests. Spencer Shepler followed with an overview of the requirements document that has been submitted for IESG consideration. Rob Thurlow concluded the scheduled presentations with an overview of the file attribute proposal that allows clients to fetch attributes individually. This generated some discussion on consideration of a set of attributes to be considered mandatory to help set client expectations and facilitate implementation. After the break, Brian Pawlowski presented an outline for supporting server failover and relocatable volumes within the protocol, similar to that supported by AFS. David Robinson described the current state of his proposal for file and record locking and Mike Eisler summarized some of the version 4 security features: v4 must support strong authentication, data integrity, and privacy. Kerberos version 5 is logical at the organization level, and a public key scheme, perhaps based on SSL cypher suites seems appropriate for the Internet. There appeared to be consensus that RPCSEC_GSS would be an appropriate security framework for NFS v4.
ONC Remote Procedure Call (oncrpc)
The oncrpc working group did not meet in Orlando
PSTN and Internet Internetworking (pint)
The pint working group meeting spent most of its time discussing the final issues concerned with the PINT protocol, but it also heard two informational presentations from TINA Consortium (which is prototyping PINT services in TINA environment) and A. Brusilovsky (who reported on the SIP-based, PINT-like implementation of Internet Call Waiting. The major agreements were 1) acceptance of L. Slutsman's proposal to support notification of the PINT Client (and, subsequently, PINT end-user) regarding the status of queued calls, which is to be supported using mainstream SIP mechanisms; and 2) agreement on the set of parameters necessary to be passed to PSTN (i.e., IN) in order to set up calls. The latter information will be communicated to ITU-T SG 11 via a liaison representative. The editors said that the final version of the draft will be ready for the last call in two months. The MIB work will be starting around that time; D. Romascanu (HUB MIB WG chair) kindly agreed to lead the work. Five PINT implementations are in progress. Several participants expressed interest in adding new PINT services (without changing the existing architecture), but relevant discussion was deferred until the March meeting, according to the charter.
RSVP Admission Policy (rap)
The working group is working on a set of 6 documents: 1 informational describing RAP framework, 5 standards' track protocol specs for COPS base protocol and its use with RSVP. Editorial and minor technical issues need to be resolved on 3 of these documents - expect to have new versions of these 3 for WG last call in early January 1999. This document set fulfills current RAP charter. Strong support within WG for a new work item to define COPS usage in DiffServ environments. This requires action from area directors and coordination with other WG chairs.
Realtime Traffic Flow Measurement (rtfm)
The rtfm working group did not meet in Orlando.
Resource Reservation Setup Protocol (rsvp)
The rsvp working group made progress in moving forward the three outstanding WG documents. There was consensus to do a working group last call on the Diagnostic spec. The group agreed to a two month timeout on Tim Moore's update to the MD5 spec to provide time to incorporate additional material. On the tunnel draft, the working group will wait for Andreas' update to meet John Wroclawski's objections.
John Wroclawski described the current understanding of RSVP, intserv, and diffserv. The "framework" document is now an ISSLL work item. This leaves to RSVP the much less clearly defined problem of the possible use of RSVP within a diffserv cloud to handle heterogeneous multicast. The relationship of this to bandwidth brokers is still unclear.
Lou Berger presented his proposed RSVP extensions for MPLS, including reliable delivery and "optimal state". The working group was asked to read Lou's draft.
There were two final presentations, one on a different approach to aggregation, and the other on SCRAPI.
Signaling Transport (sigtran)
The SIGTRAN Working Group met on Dec. 10th, with ~160 persons attending. The group reviewed its charter, which is to support the transport of packet-based PSTN signaling protocol over IP. A number of presentations were made with input on the functional and performance requirements for signaling transport, and participants were encouraged to work together on definition of the informational RFC describing these requirements. Other inputs were received on additional requirements for support of SCCP/TCAP over IP, and will be another area of discussion. On specific protocols, there was a proposal to consider RTP with Forward Error Correction as a candidate, and suggestions from the floor to consider output from the RUTS activity as well. Finally, there were reports on related work in ITU and ETSI TIPHON that identified work on H.323 over UDP that may be a useful source of information for the group.
TCP Implementation (tcpimpl)
The tcpimpl WG met on Wednesday 12/9/98 at 3:30. We first had a quick review of the status of the various WG documents. We will request IETF last call for RFC 2001.bis shortly, and intend to advance the NewReno document for publication as Experimental in parallel with it. Peter Ford led a discussion of raising the TCP initial congestion window to 2 segments in 2001.bis (targeted at proposed standard). There was unanimous acclamation for doing so. Jamshid Mahdavi gave a report on the usefulness of the WG documents for debugging TCP stacks, highlighting two new bugs, slow-start too aggressive and RTO too short, and suggesting that the Known Problems document would benefit from flagging problems as Sender or Receiver problems. These were judged as probably appropriate for an update of the known problems document, as the current version is now already with the AD. Joe Touch outlined a new algorithm for preventing bursts. The algorithm needs to be documented in an I-D before a decision is made on whether it is IETF work or research. Finally, Kevin Lahey led a short discussion on Path MTU Discovery issues. The consensus of the room was that a document on "experience with PMTU" would be useful & appropriate, with the chairs specifying that they would require a draft by January since the WG has wound down. As this document is outside the WG charter, working on it also requires the AD's approval, which was attained shortly after the meeting. We adjourned noting that we do not plan to hold any further face-to-face meetings, but that the mailing list will remain active.
TCP Over Satellite (tcpsat)
The TCP over Satellite working group met on Tuesday, Dec 8th, at the IETF-43 in Orlando for about 45 minutes. Current status was reviewed. The document "Enhancing TCP Performance Using Standard Mechanisms" advanced to BCP RFC. The document "Ongoing Research Issues in TCP over Satellite" has been updated and all sections have fairly mature draft text except for a section on Snoop and one on Interactions Between Mechanisms/Mitigations. The section on snoop will be drafted by Mark Allman and the one on interactions will be deleted since there is little material and no author for that section. Once the document has passed WG last call (expected in February) it will be submitted to the ADs for consideration as an Informational RFC. Therefore the charter of the working group has been fulfilled and the group should not be meeting at IETF-44. An announcement was made regarding the availability of a draft defining terms for a discussion on TCP performance enhancing proxies. There was a short presentation on interactions between ATM buffer management and flow control and TCP congestion control. A small amount of discussion on whether to extend the working group charter to document this issue better did not result in clear evidence that a critical mass of contributors were willing to work on a document. The chair recommended that comments on split connections, etc, be taken to the tcppep mail list and comments on performance issues and mitigations be taken to the pilc mail list.
HTTP Next Generation (httpng)
It was a bimodal result: great interest (strong consensus, probably unanimity) in desire for standards track MUX specification, interest (rough consensus) in a requirements document for something like HTTP/NG's middle layers, but little interest in and strong criticisms of standards track work at the middle and higher levels of the HTTP-NG proposal at this time. The MUX need may be addressed by the outcome of the RUTS BOF (see below).
Media Gateway Control (megaco)
The Media Gateway Control (Megaco) BOF met for 2 1/2 hours on Friday, Dec. 11. The goals of the meeting were to clarify the scope of the group's activities, achieve a common view of overall requirements, and determine a starting point for protocol definition. The meeting came to the following conclusions:
- The aim of the work should be a protocol independent of the specific applications and media bearer technologies with which it will be used, supplemented as necessary by a series of profiles
- The protocol should be capable of operation across insecure links
- The protocol should accommodate a variety of degrees of coupling between Media Gateways and Media Gateway Controllers
- The group needs to examine more use cases before it is ready to choose a starting point for protocol definition.
In summary, with respect to the stated objectives: the intended scope of the work is clearer, but will be determined by the required functions rather than the specific circumstances of use. The group made some progress toward a common view of overall requirements. The third objective was premature.
Performance Implications of Link Characteristics (pilc)
A BOF on Performance Implications of Link Characteristics was held at IETF-43 in Orlando on Dec 9. Over 265 people were in attendance. The motivation for the BoF was that TCP (and other IETF protocols) perform poorly over certain types of links. Several presenters discussed the characteristics of their links and how these characterstics affected TCP performance. Phil Karn volunteered to edit a BCP RFC describing what kind of performance links should have to carry TCP traffic and why. His proposal was enthusiastically received and many people took the microphone to support his proposal and volunteer to work on it. Several other environments were mentioned to complete the list of characteristics in the BoF charter. Christian Huitema noted that TCP was developed using a control function which assumed certain universal characteristics of the Internet which are no longer universal (e.g., symmetric bandwidth and connectivity, broadcast subnets). Discussion lasted over an hour and much of it focused on the end to end nature of the transport protocol. The first step will be to develop discussion on the mailing list of the different link characteristics, with a goal towards assessing the relative importance of addressing each, and seeing if we can attain consensus regarding which should clearly be addressed at the link layer, IP layer, or transport/end-to-end layer. The next step will involve discussion with the TSV ADs on what working group charter(s) and outputs this taxonomy then suggests. A mail list and web site are in place (http://pilc.lerc.nasa.gov/pilc).
Protocols for Negotiated QoS Multicast Communication (qosmc)
Alan Chambers (Convenor SC6/WG7) went over the evolutionary history explaining that ECTP is aimed at providing Enhanced Transport control for any networking culture, including IPv4 or IPv6, without affecting the applications above it. He explained that ECTP expects to exploit existing IP protocols, like RSVP.
Professor Kim (SC6/WG7) presented the QoS and Multicast control aspects of ECTP identifying open questions and suggesting areas where co-operation could be fruitful.
Wolfgang Fritsche (SC6/WG7) discussed Network Layer Multicast issues and suggested future co-operation on possible enhancements to IP routing protocols.
In summing up for the IETF, Scott Bradner (Co-Transport Area Director) agreed that there are known problems in the areas which had been identified and if the work is not focused, several groups could invent a variety of different solutions to the same problem. He suggested that a simple statement that isolates the problem space filled by ECTP would help him to identify the areas where the IETF might co-operate.
Participants can join the ECTP discussion list <firstname.lastname@example.org> by mailing to: <email@example.com with the message: subscribe qmc-list<originator address>
The co-chairs of the BOF agreed to ensure that this action is covered at the upcoming collaborative SC6 and SG7 session on ECTS/ECTP in January 1999 in Sydney.
Requirements for Unicast Transport/Sessions (ruts)
The goal of the RUTS BOF was to develop an understanding of the application requirements that cause numerous IETF efforts to develop their own transport and/or session protocols rather than using TCP. There were discussions regarding COPS, Radius, L2TP, HTTP-NG, SIP, NFSv4, SS7, IP telephony, BGP, and DNS, followed by Steve Bellovin discussing why removing TCP's 3-way handshake in an effort to attain lighter-weight transactions introduces security problems. The main requirements expressed were: multiple concurrent, congestion-observing connections between two hosts; quick session establishment; visibility into network conditions; control over reliability and retransmission; lightweight server state; record marking; failover; and per-message priority. A mailing list for further discussion of the requirements has been set up as firstname.lastname@example.org, subscription via sending email to email@example.com with "subscribe" as the subject.