[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
UDLR minutes
here follows the draft minutes.
please send comments to the list.
Walid
---------------
Minutes of the UDLR BOF
San Jose IETF, Dec 10th, 96
by Walid Dabbous (thanks to Jean Bolot for taking notes)
The BOF was structured in three parts, denoted A, B, and C
Part A
- Agenda bashing
- General presentation
- Very quick description of the problem and recapitulation of Montreal
BOF
Part B
Made up of four presentations:
Presentation 1 (by Emmanuel Duros, INRIA)
- Experimentation at INRIA with modified RIP.
The INRIA solution is based on modifying routing daemons. Receivers
detect dynamically the presence of feeds and send routing messages to
them via regular connections.
There were a question from the AD whether there might be a bootstrapping
problem in the general case, i.e if there were more than one
unidirectional link. This is the general case, and will not be addressed
by the UDLR group.
Jon Crowcroft: one way around that is to use OSPF instead of distance
vector-like RIP. Or let the receiver mcast the information.
Steve Deering: also there is an implosion problem because if there are N
receivers, then N flows on the return path would overload links back and
the source as well.
- Modifications to DVMRP:
Regarding implementation of DVMRP, Steve Deering pointed to Tom
Puisetiari's draft (available for 6 months).
INRIA is currently implementing the modification to DVMRP but these
modifications are not working yet.
- DVB interface:
Description of the DVB interface at INRIA and of the European MERCI
testbed.
Presentation 2 (by Yongguang Zhang, Hugues research lab)
Defines 2 UDLR models:
1. udlr on top of bidirectional underlying network (wired Internet)
2. bidirectional islands connected by udlr
Mentions bootstrapping problem in general case, seems like no answer
available.
Hugues' solution to pb 1, is based on the use of tunneling. They
implemented it in linux. Hugues says their problem translates into the
asymetric link metric.
This is true in general, unidirectional links are just a special case of
non symmetric links, with one metric being infinity.
Short discussion about whether INRIA and Hughes solution allow
autodiscovery.
Presentation 3 (by Akihiro Tosaka, the WIDE project)
Their approach is not based on dynamic routing.
Solutions:
- For the backbone application (equivalent to INRIA's subnetwork
access), the solution is similar to Hughes' (called VIF, Virtual
InterFace message). The unidirectional link is hiden at level 2 by using
a VIF. Does not require a change in RIP current implementation. OSPF in
debugging mode
- For the home application, the solution is based on Network Address
Translation (NAT) (select return path depending on application).
Problems of the approach:
Scalability
Lack of transparency
MAC address of satellite network.
Ongoing work:
- VIF: finish OSPF, experiments in larger environments
- NAT: implementation and evaluation
- Support of multicast
Presentation 4 (by Scott Corson, University of Maryland)
Multicast tree construction in directed networks
A presentation about ongoing theory work on the more general case of
links with non symmetric metrics.
They assume an underlying routing protocol (so udlr overlay, same as
case 1 above)
In fact, amounts to adding/deleting senders and receivers using a CBT
like architecture (with core/RV points)
Part C
Part C was a discussion, the goal of which was to decide:
do we need a WG, and if so what are the goals?
Walid presented the proposed WG charter.
(See below)
Emphasizes that WG would not consider impact on higher level (RSVP)
Discussion:
- Somebody says that the work should not be restricted to satellite
unidir links, but to unidir links in general
- Steve D: VIF/Hughes does not need modification to routing protocols
whereas INRIA does, so what is the advantage of INRIA's approach?
- Walid D: At this point, no claim from INRIA about which one is
better. Scalability and transparency issues should be considered in the
comparison.
- Steve D: Regardless of solution, need to look at routing specific
solutions to non-symm links. So maybe the WG should be asymetric link
routing, not unidir link routing.
- Joel says no, need to stay focused (and if everything's done in 2
meetings, then fine)
- Question about ADSL. What's the relation with IP over other non
symmetric links like ip over cable, issll?
- If the answer is a tunnel, then please take one that already exsits,
and that works. Joel: Note that what we are talking here is a control
tunnel.
- Question about link state. No work yet on LS routing, yet needs to be
looked at (for commercial reasons) and suspects there might be pbs.
- Steve D: how much do you expect to learn from implementation of your
schemes? He suspects whatever implementation results are found by
Hughes/INRIA will not help in deciding which scheme to pick.
- Walid D: back to having a focused WG. And shows goals/milestones (see
proposed charter below).
- Again mention ip over cable etc, wgs that look like they have stuff in
common. In fact, does not seem like we know what the cable model is.
- Joel: all milestones need some kind of output (draft or whatever) for
the charter.
- Walid and Yongguang are willing to co-chair the group.
- Joel says continue on mailing list.
-------------------below is the proposed charter
DRAFT UniDirectional Link Routing (udlr) WG Charter
Co-Chairs
* W. Dabbous <dabbous@sophia.inria.fr>
* Y. Zhang <ygz@isl.hrl.hac.com>
Routing Area Director:
* Joel Halpern <jhalpern@newbridge.com>
Mailing List Information
* General Discussion:udlr@sophia.inria.fr
* To Subscribe: udlr-request@sophia.inria.fr
* Archive: ftp://zenon.inria.fr/udlr/archive.txt
http://www.wins.hrl.com/udlr/archive.txt (mirror)
Description of Working Group
High bandwidth, unidirectional transmission to low cost, receiver-only
hardware is becoming an emerging network fabric, e.g. broadcast
satellite links or some cable links.
Two cases for unidirectional links support may be envisaged:
1. unidirectional links on top of bidirectional underlying network
(wired Internet)
2. bidirectional islands connected via unidirectional links. In both
cases, the integration of unidirectional links may require changes to
the routing protocols in order to preserve dynamic routing across these
links. A short term solution (i.e. to solve the first case) is to adopt
current protocols with possible modifications. A long term solution
(i.e. for the second case) is to propose, design and implement protocols
that remove assumed link symmetry (e.g. by supporting 2-way metrics).
There have been several proposed approaches for the short term case.
The first is based on the modification of the common routing protocols
(RIP, OSPF, DVMRP) in order to support unidirectional links. The second
is to add a layer between the network interface and the routing software
to emulate bi-directional links through tunnels.
The purpose of the UDLR Working Group, therefore, is to study these
approaches and suggest a short term solution to provide dynamic routing
(including multicast) in the presence of unidirectional links.
Goals and Milestones
April 97
Meet at IETF. Finish discussion of the general solution.
Post the description of the proposed solutions and
the issues for the short term solution as I-D(s).
August 97
Post the description of the short term solution as an I-D.
Preliminary experiments results of the short term solution.
December 97
Post description of the recommended short term solution as an
informational RFC.
End WG