APCCIRN-033
APCCIRN-033
Minutes from the IEPG meeting at NIKHEF
Amsterdam April 26, 1993
Bernhard Stockman
Participants
Guy Almes ANS <almes@ans.net>
Peter Dawe PIPEX <peter@pipex.net>
Elise Gerich Merit <epg@merit.edu>
Daniel Karrenberg RIPE NCC <daniel.karrenberg@ripe.net>
Peter Lothberg EBONE <roll@bsd.stupi.se>
Olivier Martin CERN <martin@cearn.cern.ch>
Jun Murai WIDE <jun@wide.ad.jp>
Bernhard Stockman EBONE <boss@ebone.net>
Claudio Topolcic FEPG <topolcic@cnri.reston.va.us >
1. Opening
1.1 Welcome, appointment of scribe
EG and BS welcomed the participants.
EG volunteered to chair the meeting and BS volunteered to take the
minutes.
1.2 Approval of agenda
The agenda was approved after two additions:
- Presentation of European GIX by PD, added at agenda point 3.
- Future meetings planning by BS, added at agenda point 6.
1.3 Approval of the previous minutes from the IEPG/GIX meeting,
March 31, 1993 Columbus, Ohio, USA.
Previous minutes were updated with proposed changes and was
approved.
2. Global Link and Routing Coordination
After April 1994 the NSFNET backbone as we know it might no longer
carry transit traffic. This includes all transit traffic and all
primary traffic within NSFnet.
Networks dependent on NSFnet for their traffic should be aware of
the consequences. Communities reachable via NSFnet such as part of
Canada, Brazil, Argentina, Equador, Central America will be
affected. US midlevel networks will obtain global connectivity as a
result of the fourth part of the anticipated NSF solicitation. This
might include using backbone providers or direct connectivity to the
GIX, or other options they might propose.
A transition plan has to be presented by the bidders and to be
worked out with the current NSFnet. The time-frame is unknown but
probably around May 94.
The meeting agreed on a recommendation that the only short term
solution is the GIX. Guy Almes proposed the following text:
"If every network operator arranges that, either directly or
indirectly, they can both exchange packets and exchange routes at
the Washington GIX, then we can expect that no global
discontinuity will occur upon the April 1994 NSFnet transition."
The statement will be discussed on email and sent to the IEPG list
for approval and when finalized sent to CCIRN as a recommendation.
3. GIX Status
3.1 Presentation of European CIX by PD
Internet is today an anarchy. The changes over the next year are
such that people are uncertain on its status. There is a need for a
global authority to regulate plans of the global internet. There
are today several bodies like RIPE, RARE, CCIRN/IEPG, each of which
covers a subset but not one that covers everything so one has to go
to all these meetings.
A group could attain authority by achieving and having the control
over the GIX. PD proposes by this that IEPG should be given this
authority
There is a need for regulation of Neutral Interconnects, *IX's. If
not neutral there is a risk of abuse. Issues of charging models has
to be addressed. There is also a need for regulation of peering on
various levels.
With the implementation of the European CIX the inter-CIX
connectivity has to be 100 percent paid by the Europeans which is a
seen as a problem.
PD proposes that IEPG is asked to be become such authority under
supervision of IAB and that IEPG takes ownership of the GIX and uses
that to assert its authority of global interconnectivity. The
proposals could take the form of RFC's.
The proposals were discussed.
A centralized authority may create a potential threat to some
network providers which may give loss of support for the GIX and by
that be counterproductive. We may achieve the same fairness in
coordination without enforcing authority.
Bilateral peering agreements are sometimes not fair taking into
account already existing agreements. If new providers comes on, this
may lead to unfairness if not already there. We got the to get the
regulation to assure that some does not abuse.
Regulations by authority may be to go to far and could be achieved
by recommendations. For this to work we need that providers
documents their needs so that the system is possible to
troubleshoot.
Conclusions:
There was much discussion on the difficulty of achieving PD's vision
without inappropriately seeking more power than the community may be
willing to grant. PD will submit his proposal to the IEPG list for
further discussions.
We could stimulate the forming of a group to set the regulatory
environment to which providers can subscribe similar to a trade
association.
3.2 GIX status
In total 6 routers installed at GIX and the Euro-RS is up and
running. RIPE NCC is working on implementations to make it
functional. The RS will initially be used as primary routing source
for some networks. The plan is currently one month behind schedule.
Merit RS is one month behind schedule. Merit is developing a program
that will take the NSFnet database primary AS's to be used as info
and base for the Merit RS.
Asia:
AP-NIC welcomes the RS initiative. At a January meeting it was
decided to continue the work in this area. JP-NIC has been formed
and got the necessary resources.
At the question whether statistics will be made available, it was
pointed out that statistical gathering in a commercial environment
may have legal implications which must be addressed.
The ripe-82 paper explains the routing concept for the GIX. The
paper is available via anonymous ftp from the RIPE document archive,
archive.ripe.net.
Multiple Interconnected GIX's
There will be multiple points of gravity in the Internet whether we
like it or not. Talks about multiple GIX are already happening and
the issue has to be addressed. Multiple routing cores and cost
sharing are important issues to be addressed. In essence we are
already facing the problems of multiple GIX's as there are multiple
routing authorities.
We need a paper discussing whether it is useful to talk about more
than one GIX. There may be other efforts trying to build more GIX's.
There can however never be more than one final top-level destination
for resolving routes not resolvable at any, routingwise, lower
instance.
It will be necessary to establish a top level registry as we need
one registry describing how providers talk to each other. By that
the routing topology is defined.
What we could try to define is where peer and transit occur.
At this point PL presented his Internet eXchange Facility (IXF)
model. A paper describing the IXF model is under preparation and
will be distributed to the IEPG list for further discussion.
Some remarks from the presentation:
The GIX functionality can be divided in two parts:
- The GIX packet exchange part (PX)
- The GIX routing registry system (RR)
Technical Requirements for a PX
- PX organizations are independent of each other
- PX organizations collaborate with the RR
- PX organizations can not be part of any service provider.
- A minimal PX has two service providers connected
- No usage restrictions are enforced by the PX's as such
The group agreed that multiple PX are likely to happen and must be
addressed. PX can interconnect via service providers providing
transit or several SP's joining for shared resource.
Registry part
Multiple PX's can only work if there exist a coordinated system of
RR's. We have to guarantee that the RR's are actually used by the
SP's connected to the PX's. This may be self-regulatory as SP's not
registered will not get any packets.
We can not get all to buy in to this but we can try to align ongoing
efforts to create such a global information. It will be necessary
to find as neutral as possible organization to provide the RR
functionality.
We will need a global database. For the Global RR system to work
there is a need for a agreed format for this global database. Work
is going on within RIPE and MERIT with a Policy Description Language
(PDL) as was presented at the previous IEPG meeting.
Conclusions:
On PX's
Multiple PX will exist. Peters paper will be discussed at the
mailing list distinguishing between the technical and administrative
parts.
On RR's
There is a need for distributed RR system. Necessary with a RR-PDL
to make this work. The RR-PDL current proposals are not necessary
what we want in the end but collaboration among the people involved
is encouraged.
4. Scaling and Forecasting
The Internet growth under-estimated. European growth is today
faster than within the US.
What can we measure:
It would be useful to have the CIDR core members publish there
routing table size statistics. CIDR core router resource
consumption statistics are needed. One metric would be historically
give memory usage vs network numbers to be able to forecast resource
utilization.
This can probably not be achieved on a voluntary basis. It will be
necessary to find resources to accomplish this.
CCIRN have asked feedback on recommendations for critical projects
to be funded by CCIRN member organizations and this project seems to
be a good candidate.
EG will put together a draft project proposal to be sent to the IEPG
mail list for discussion and, when finalized, sent to CCIRN.
5. Operational Impacts of IP Next Generation Transitions
We should start to worry about the need for renumbering. We need
tools for dynamic host configuration to make renumbering
transparent. This could be another candidate proposal for a project
to CCIRN. It was noted that a request for a renumbering scheme
should not be made public premature as this could create a lot of
worry. We need the tools to be there first.
Conclusion:
Whatever comes out look at number-space. We also need to look how
to build in IXF's to make the whole architecture scale.
There is not an immediate need to act now. The issue will probably
be addressed within IETF and IEPG could await that to bring
additional value to the table. The current process is possible to
influence in the IETF.
To minimize the CIDR entropy of CIDR renumbering is possible.
Automatic renumbering could be done via IPX's.
EG will draft a proposal to be discussed at the IEPG mailing list.
6. Any Other Business
Announcement of IDPR project.
The project have received policy statements from NSFnet NASA and
TWBnet. IDPR has thus initiated a pilot project to evaluate the
concept.
IEPG meetings
It was agreed that IEPG meetings shall happen in conjunction with
other major global events such IETF and INET. We should be in
charge of our own meeting facilities not to come in conflicts with
other meeting organizers. Sunday meetings are preferable.
INET's have so far been starting on Tuesday so, when piggybacking on
INET conferences Mondays are also possible. As there tends to be
more Asians being able to attend INET conferences then IETF's the
priority is INET's and to meet the day prior to the conference
start.
The week after the INET-93 there is a proposed joint CCIRN/IEPG
meeting. The meeting place is already decided to be Bodega Bay
outside of San Francisco.
The meeting saw it as useful to have two joint sessions, one on
Monday afternoon and the second on Wednesday morning. The IEPG
chairs will send a proposal to the CCIRN.
7. Closing
EG and BS thanked the attendees and the host (NIKHEF) for a
productive and well organized meeting.