Minutes from the last meeting - blush!

[Duncan Rogerson <]

Folks,

I just realised that after being keen and typing these minutes
up straight after the London EOF meeting in November, I never
circulated them!  Many apologies - here they are, for those that
haven't already started travelling to Amsterdam!

Dunc

---




European Operators Forum

Minutes of the 4th EOF Meeting                                Duncan Rogerson

Held at Telehouse, London, on Friday, 25th November 1994                 ULCC
                                                           27th November 1994

_____________________________________________________________________________



Present :


Judith Blair            PIPEX           judithb@localhost               (JB)
Graca Carvalho          RCCN            graca@localhost                  (GC)
Sean Doran              Sprint ICP      smd@localhost                     (SD)
Harvard Eidnes          NORDUnet        Havard.Eidnes@localhost   (HE)
Jorgen Ericsson         Tele2           eri@localhost                    (JE)
Frode Greisen           EBONE           Frode.Greisen@localhost          (FG)
Peter Lothberg          EBONE           roll@localhost                   (PL)
Keith Mitchell          PIPEX           keith@localhost                 (KM)
Marc Pichon             TRANSPAC/RAIN   pichon@localhost                  (MP)
Duncan Rogerson         ULCC/JANET      D.Rogerson@localhost          (DR)
Willem van der Scheun   SARA            scheun@localhost                  (WS)
Fergal Suipeil          HEAnet          fergal@localhost           (FS)
Nigel Titley            BT              Nigel.Titley@localhost             (NT)
Bernard Tuy             RENATER         Bernard.Tuy@localhost          (BT)
____________________________________________________________________________



1.  Opening

Peter Lothberg welcomed everyone to  the  4th  EOF  meeting.
Duncan Rogerson volunteered to take the minutes.

Apologies were received from :

Per Gregers Bilse, EUnet; Stephan Biesbroeck, BELNET; Chris-
tophe  Chaillot,  France  Telecom;  Peter  Dawe,  PIPEX; Luc
Dierckx, INNET; EUnet GB; Hakan  Hansson,  Unisource;  Kevin
Hoadley, ULCC/JANET; Daniel Karrenberg, RIPE NCC; Lars-Johan
Liman, SUNET; Mike Norris,  HEAnet;  Arnold  Nipper,  XLINK;
Jean-Michel  Plance,  Oleane;  Tim Streater, DANTE; Wilfried
Woeber, ACOnet;

The agenda, as circulated by KM, was approved.


2.  Approval of Minutes, Review of Action List

The minutes of the 3rd EOF meeting were approved.

The main action brought forward from the 3rd meeting was the
work   on   drafting  an  agreement  form  for  interconnect
partners.  PL had circulated example agreements to  the  EOF
mailing  list,  and  will continue to coordinate the work to
draft a generic interconnect agreement.

This work was discussed under a later agenda point.

      Action 4-1 : Peter Lothberg to coordinate interconnect
agreement work


3.  European Interconnect Points - Updates


3.1.  London INternet eXchange - LINX (KM)

+ Physical configuration is currently an ethernet hub
+ Connections to BTnet and JANET LINX routers are E1 lines
+ PIPEX connects via ethernet to their Telehouse POP
+ BTnet, JANET and PIPEX are all connected and peering across the LINX
+ Demon Internet and EUnet GB due to connect soon
+ Documentation and MOU is available via WWW : http://www.linx/net/
+ No transit for small providers - must have own non-UK connectivity
to connect to the LINX
+ Only UK providers are connected at the moment
+ LINX network number is from Stockholm D-GIX block
+ Switched ethernet planned to replace hub
+ JANET shortly to install a POP at Telehouse, connected via 10Mbps
SMDS and E2 link to ULCC.  E1 to be replaced by ethernet connection

The interconnect facilities are funded by contributions from
the partners.  Interconnect management is by a group formed
by representatives from the partners.

PL asked is this model could be extended to other common
areas in Internet network service provision - eg namespace
management.  In the UK, the namespace is managed by a
cooperative group of representatives from all the UK service
providers.


3.2.  Paris F-GIX (BT)

+ Currently implemented as a managed, transit AS (AS3256)
+ Oleane and Renater connected
+ PHYnet (French High Energy Physics community) to connect soon
+ Service providers to be allowed to connect direct to the
transit media, and peer with AS3256
+ PIPEX to be directly connected soon
+ EUnet France intend to connect
+ There is an MOU, in French.  BT will translate and circulate the MOU

It was noted that the transit AS model is how the current
NSFnet backbone is implemented, and means that networks con-
necting to the managed AS3256 cannot have their own policies
to other F-GIX connected networks.

AS3256 could provide transit to EBONE, with the permission
of EBONE.


3.3.  Stockholm D-GIX (PL)

+ No major changes since EOF-3
+ Connectees include SWIPnet, TIPnet, SUNET, Transpac, DATAnet, NORDUnet,
EBONE and PIPEX,  + others
+ Physical configuration is two Cisco Catalyst switches
+ Each service provider has own port
+ Provider access routers are mostly Cisco 2500
+ Route server unlikely for at least 1 year
+ Government funds hardware and installation/operation
+ Service providers must have peerings with at least two others to connect
+ There are plans for regional interconnect points within Sweden

PL also gave a short presentation on how the view of the D-
GIX and the outside world from within SUNET.


3.4.  US Infrastructure (SD/PL)

+ Chicago NAP, based an ATM, not yet operational because of ATM-related problems
+ New York NAP, based on FDDI, is operational.  Plans for a DEC FDDI switch
+ MCI, ANS and SprintLink peering over NY NAP
+ San Francisco NAP, based on ATM, operational, but suffering from
throughput problems due to ATM issues.  MCI and ANS were peering over the
NAP, but now use a private FDDI.
+ A DMZ FDDI interconnect is developing at NASA Ames, independently from
the NAP project.  Named MAE West, most of the major US players are expected
to connect.


3.5.  General Issues

+ Advertising the interconnect network number

A route to the interconnect number should only be carried in
a provider's IGP, and not announced to the rest of the
Internet.  Access to a provider's router, if its connection
to the interconnect failed, should be by using the address
of another interface, perhaps a loopback interface.  These
addresses should be registered in the RIPE database, in the
inet-rtr: object.

+ Multicast on the interconnect media

The F-GIX and LINX do not yet carry MBONE traffic, although
the partners of each interconnect are discussing multicast
issues.  Some experiments have been carried out on the
Stockholm D-GIX, with SUNET providing the MBONE feed.

Cisco's PIM implementation was discussed, and several prob-
lems noted - pruning and joining issues for people without
tunnels, how to get DVMRP routes through PIM to the MBONE.
These are going to be addressed by Cisco at the next IETF
meeting.

Raw multicast could be run over the interconnect media,
allowing service providers to pick up and dump traffic as
required.  This happens on MAE East, although it suffers
from loss problems - peak MBONE packets can reach 2000
packets/second, when 2 video feeds are active.  Multicast
routing loops could also be a problem.


4.  Interconnect Agreements

PL had circulated sample agreements, work to put together a
generic agreement is ongoing.

It was noted that there may not be a need for complex agree-
ments in Europe, at least in the short term.  The existing
interconnect points were operating without problems, includ-
ing MAE East in the US.  Any problems are physical implemen-
tation limits, and not higher level disagreements amongst
partners.

A generic agreement may be hard to write - partners at dif-
ferent interconnects may have different requirements.  Ulti-
mately, the interconnect partners have to agree what to
agree.

Interconnect implementation agreements should be separate
from peering/traffic agreements.

The recurring problem of what distinguishes a service pro-
vider from a customer was discussed.  The meeting concluded
that a set of rules was emerging from experience, and that
these rules could be documented.

It was noted that private connections between providers
should be used before public interconnects, although the
public interconnect could be used to backup private links.


5.  EOF Chair, IEPG European Co-chair for 1995

Peter Lothberg was elected the chair of the EOF, and Harvard
Eidnes the European Co-chair of the IEPG.  Both would act as
stand-in for the other position.

It was suggested that both positions be reviewed yearly.


6.  EOF Charter

The EOF Charter stands, as approved at EOF-3 in Lisboa.


7.  Impact of US Changes on European Internet Connectivity


7.1.  CIX

Following the CIX decision not to route packets for non-CIX
members, there are a small number of non-CIX members, who
have no connectivity to CIX connected networks.


7.2.  NSFnet

As soon as alternative connectivity is arranged for all NSF
sponsored networks, the NSFnet backbone will be switched
off.  NSF are hoping to do this as soon as possible, and by
1st April, 1995, at the latest.

The NSF funding changes also mean that when the current NSF
international connectivity agreements expire, new arrange-
ments will have to be found for US-European connectivity.
US service providers are expected to start operating in some
European cities in the next year, which may provide a solu-
tion.

Asia-Pacific connectivity from Europe is unsure at the
moment, following the NSF changes.  This issue is under dis-
cussion within the IEPG.  Countries/networks buying their US
connectivity via a US service provider should be able to
route to Asia-Pacific networks via the US provider's NAP
connection.

The meeting concluded that US carriers should be invited to
connect to European interconnects.  It should also be sug-
gested to the IEPG that an exchange point is implemtented
for each continent.


8.  CIDR and Routing Issues

A general discussion on CIDR and routing followed, covering
the following points :

+ CIDR deployment is too slow

+ EOF should encourage service providers to have customers use
addressing from the provider's block.  Customers should be encouraged
to renumber when changing provider, and to use service provider addressing
instead of `old', pre-CIDR allocation network numbers.

+ A common policy for all European service providers is needed
to help encourage organisations to renumber.  Unless all providers
use the same policy on renumbering, there are competition issues -
an organisation may pick the service provider who is the least
strict about renumbering.

+ Class B subnets should not be routed - this is only useful for
a multi-homed organisation, network numbers from the provider's
block be used instead.

+ Reducing route flaps : don't use dyanmic routing over dialup
connections.  Sprint are asking Cisco for a configuration option
to prevent a route being withdrawn from the BGP table when a link
goes down.  This also has the advantage that traffic always routes
via the correct provider, and does not default to the provider routing
a block of numbers when a specific route goes away.


9.  Next Meeting, Closing

The 5th EOF meeting will be held at 09:00 on the 25th Janu-
ary, 1995.  This is the morning of the first day of the next
RIPE meeting, being help at CWI in Amsterdam.

Peter Lothberg thanked everyone for their contributions, and
PIPEX for arranging the meeting and providing lunch, then
closed the meeting.


10.  Action Summary

Action 4-1 : Peter Lothberg to coordinate interconnect
agreement work