MBone Working Group

Minutes of the RIPE-30 Mbone WG meeting

Chair: Thomas Telkamp (replacing Kurt Kayser who was unable to attend)
Scribe: Felix Kugler
Attendees:  41

0. Administrativa

- no additions to the agenda proposed
- no comments to the minutes of the last meeting (RIPE-29)
- no review of the action list (there were no actions)

1. Status of the European Mbone

1.1 current topology

There were a lot of changes in the past four months since Dante has
introduced a number of new Mrouters. An updated map is available from 
http://www.ibbnet.nl/mbone/mbone-eu.ps. Leaf nodes have been omitted
now to clarify things.

Michael Behringer talked in detail about the recent topology
changes. Dante has setup Unix based Mbone routers in their
POPs and aligned the primary tunnels to the TEN-34 topology. NRN
access links are now free of permanent tunnels to other NRNs. It is
now fairly easy to install new tunnels if needed. A detailed map
covering the TEN-34 part of European's Mbone is available at Dante:
http://www.dante.net/~nmc/mbone/Mbone_ten34.ps.

The transition of the primary tunnels mostly worked as scheduled. Due
to a relocation change of the DE-POP, the two tunnels to Renater and
CESNET are still waiting for their final arrangement. Installation of
the backup tunnels will follow soon.
 
1.2 Software versions

- mrouted3.9b3 seems to run stable

- Cisco IOS: 
  generally use 11.2 or latest 11.3 releases.
  Actual releases in April 98 were IOS 11.2.13 and IOS 11.3.2.
  Use of 11.3 version is required in redundant setups (i.e. if you
  have loops) due to a forwarding bug described on 
  http://sandbox.parc.xerox.com/fenner/mcast/flooding.html.

  Wilfried Woeber doubts whether the forwarding problem is really
  fixed in IOS 11.3 releases. He reports another weired phaenomenon
  where Cisco's do not prune at all in redundant configurations and
  even keep on flooding traffic when the receiving mrouted has died.

  Wilfried Woeber asks if there is anybody in Europe who knows details
  about Ciscos IOS versions (included features, known bugs etc). So
  far, such questions can only be answered by a few "gurus" in the
  US. They all are overloaded with work and often answer only to
  people they know personally. 

- ICMP errors messages are sent by a few TCP/IP implementations when they
  see native multicast traffic though this is forbidden according to
  RFCs 1112, 1122, 1812. A list of known buggy implementations is
  available on http://sandbox.parc.xerox.com/fenner/mcast/icmp.html.

1.3 Pruning

Several European networks are still non pruning. The general feeling
is that this situation is caused by configuration errors rather than
by too "old" software. Michael Behringer points out that there is an 
Mbone version monitor running in Cambridge which allows to find outdated
Mrouters easily (http://www.cl.cam.ac.uk:80/cgi-bin/mr-vers).

Downstream non-pruners usually can be found using "mrdstat" for
mrouted (http://www.switch.ch/lan/mbone/tools/mrdstat) or "mrtg" for
routers (http://ee-staff.ethz.ch/~oetiker/webtools/mrtg/mrtg.html).

Wilfried Woeber suggest to start actions against non-pruners
now. There is no easy way to impose restrictions on misbehaving
downstream sites, however. Cutting off multicast tunnels is an
unfriendly way to handle the problem, and setting very low rate limits
on tunnels to non-pruners does not help at all to reduce the bandwidth
consumption on the upstream side.

Action (30.1) on Wilfried Woeber and Thomas Telkamp: to start tracking
down non-pruning networks on the ams-mr.att-unisource.net router.

1.4 Various

Mbone Instabilities

The European Mbone is currently running unstable. Unix-based Mrouters
often freeze for several minutes while at the same time multicast
traffic is very high. Though the problem is not yet fully understood,
it seems clearly related to loop topologies where Cisco-based routers
and Solaris-based mrouteds are involved (ex: the loop
ams-mr.att-unisource.net - ch-ws.ten-34.net - mbone-cern.switch.ch). 

For more info about the forwarding loop problem see
http://sandbox.parc.xerox.com/fenner/mcast/flooding.html. There is a
kernel patch available for Solaris 2.5.1, a similar fix for Solaris
2.6 is under test right now.

Admin scopingWhat could be done with admin scoping in Europe ? Should the European
Mbone be taken as one region ? Which addresses could be used ?
Appearently, the audience was not fit for a discussion and no 
conclusion could be achieved.

2. MPEG-L3 Codecs

Martin Sieler from Fraunhofer Institut f|r Integrierte Schaltungen
gave an overview over the latest developments in the area of compressed
audio signals. The talk also included a number of impressive
demonstrations, namely a comparison of two 64kbps coded audio streams,
ISDN (A-law) and MPEG-2AAC, which proved how much progress was
achieved in the past few years. 

MPEG-2AAC (Advanced Audio Compression) is a new, flexible coding
scheme which can be used in low bandwidth environments (two-channel
audio with 16kbps) as well as for broadcast quality sound (two-channel
audio with 128kbps). MPEG-2AAC supports sampling rates from 8..96kHz,
1..48 channels, 1..6 bit/sample. Typical applications include SW/MW
digital radio, Internet audio, solid state recorders.

Detailed information about audio compression schemes, downloadable
software and demonstration audio files are available from
http://www.iis.fhg.de/amm. 

3. AMS-IX multicast experiment

Thomas Telkamp informed about the plans to introduce native multicast
on the AMS-IX in Amsterdam. 

The idea is to use tunnels from the ISPs to the AMS-IX access routers,
but to use native multicast on the IX itself, reducing the load
there considerably. Tests have shown that the Catalyst switches are
able to cope with 30Mbps multicast traffic without showing much CPU 
load.

A number of topics like correct pruning on the LAN, MBGP-IGP
redistribution and others will have to be looked at carefully. The
migration is planned in several phases during this summer. The final
report is expected by September '98.