APCCIRN-018
APCCIRN-018
1993.02.03
IEPG
MINUTES OF IEPG MEETING
held at
Quality on Capitol, Washington DC
0930, Sunday 15 Nov 92
The fifth meeting of the IEPG was called to order at 9:30 am by
Bernhard Stockman; co-chair Elise Gerich arrived shortly after. As
well as IEPG members, representatives of network providers had been
invited to the meeting. Attending were:
George Abe Infonet abe@infonet.com
Vikas Aggarawal JVNCnet vikas@jvnc.net
Guy Almes FARnet almes@ans.net
Tony Bates ULCC t.bates@noc.ulcc.ac.uk
Nevil Brownlee Tuia nevil@aukuni.ac.nz
Brian Carpenter CERN brian@dxcern.cern.ch
Bob Collett SPRINT rcollett@icp.net
John Curran NEARnet jcurran@nic.near.net
Peter Dawe PIPEX peter@pipex.co.uk
Peter Ford LANL peter@lanl.gov
Gopi Garge ERnet-India gopi@ece.iisc.ernet.in
Elise Gerich Merit/NSFnet epg@merit.edu
Masayoshi Gohara NACSIS/SINET mg@sinet.ad.jp
Tony Hain ESnet/LLNL hain@es.net
Juha Heinanen Telecom Finland jh@datanet.tele.fi
Daniel Karrenberg RIPE NCC karrenberg@ripe.net
Jong Yeol Kim Korea Telecom kimjy@ring.kotel.co.kr
Peter Lothberg EBONE NOC roll@stupi.se
Jun Matsukata ISAS, NACSIS/SINET jm@eng.isas.ac.jp
Jun Murai Wide Project/Japan jun@wide.ad.jp
Andrew Partan UUNET asp@uunet.uu.net
William L. Schrader PSI wls@psi.com
Erik Sherk SURAnet sherk@sura.net
Brian Shiflett SPRINT brian@icp.net
Joo Y. Song HANA jysong@ring.kotel.co.kr
Bernhard Stockman EBONE boss@ebone.net
Fumio Teraoko SonyCSL/Wide tera@csl.sony.co.jp
Marten Terpstra RIPE NCC marten@ripe.net
Claudio Topolcic CNRI/FEPG topolcic@cnri.reston.va.us
Jack Waters SURAnet waters@sura.net
Wengyik Yeong PSI yeong@psi.com
1. GIX Discussion with Invited Guests
----------------------------------
Bernhard introduced this topic by calling attention to two
documents:
* Draft Proposal for Global Internet Connectivity; IEPG Working
Document Guy Almes, Peter Ford and Peter Lothberg; June 92
* Internet Routing in a Multi Provider, Multi Path Open Environment
Tony Bates, Daniel Karrenberg, Peter Lothberg, Bernhard Stockman
and Marten Terpstra; 12 Nov 92.
The first explains the GIX concept and is recommended reading. The
second proposes a pilot GIX implementation, to be running by 15 Mar
92.
1.1 The GIX Pilot Proposal
----------------------
Peter Lothberg reviewed the pilot proposal. The aim is to have a
LAN here in Washington to which all network providers can connect.
They will register their routes in a database which will be used to
maintain a Route Server on the LAN; the database will not be tied to
any particular policy. The Route Server will receive route
announcements for all networks and filter them using the database.
In addition individual nets may pair across the LAN, which allows
for exceptions to the general scheme.
Daniel Karrenberg provided further explanation, pointing out that
the aim of the proposal is to make sure that the Route Server has a
consistent view of routes into Europe, which it can distribute back
to any router it chooses to. In time we should implement similar
Route Servers for the North America and Asia/Pacific regions.
There was considerable discussion, which brought out the following
points:
* All network providers have to connect to the GIX. In this way
there is no connection hierarchy (which could impose routing
policy).
* The GIX concept does not depend on the choice of LAN medium. A
wide-area ethernet is proposed for the pilot simply because it is
readily available now. A point-to-point network connecting
routers in various parts of the world may be a better way sometime
in the future.
* The RIPE NCC has an effort underway to collect all the necessary
routing information and store it in their database for European
networks. This is expected to be sufficient for the pilot, but
the other regions will have to set up their own Route Servers.
The Route Servers will have to interact to provide each other with
stable routes between regions.
Overall it was AGREED that the pilot implementation was well
worthwhile, and should proceed with all possible speed so as to be
able to report results to the March 93 IETF meeting. The U.S.
network providers will help by providing the RIPE NCC with data
about routes to their networks.
1.2 The Washington Proto-GIX Implementation
---------------------------------------
Andrew Partan from UUNET made a presentation about PSI, Alternet and
Sprintlink's Metropolitan Area Ethernet (MAE) implementation using
MFS's 10 Mbps Ethernet in the Washington area - also known as
MAE-East. MFS provides each connected site with an ethernet AUI
cable connection; to the sites the system appears to be an ethernet
LAN. Each network provider will attach a router to the LAN; n**2
routing is used between the routers. All network providers are
welcome to join the MAE-East.
MAE-East appears to be a very good place to put the pilot route
server for the European networks. A lively discussion ensued,
covering the technical details of the MFS service, and especially
the implementation and location of the pilot route server. Among
the points of interest were:
* The LAN is a broadcast medium. It could be hard to determine the
source of bad packets appearing on it. In the long term someone
will have to provide proper operations support. Short term the
proposal should include procedures for handling faults.
* The CIX group supports the MAE-East project, but doesn't want to
influence it directly.
* By next year MFS should be able to provide connections at speeds
higher than 10 Mbps. Other communications suppliers have products
similar to MFS.
* We need to think about a funding model for the GIX. Long term a
distributed interconnection may well be a better way to connect
routers to the GIX, since it minimizes the connection cost for
each network provider. Short term the Washington MAE is the
simplest way of providing connectivity to the route server; it is
therefore the best choice for this pilot project.
* The MAE can provide traffic capacity between providers. All
routers must use the route server as their 'general case' source
of routing information. Pairs of networks needing greater traffic
capacity may, however, make direct interconnects between
themselves, and use 'special case' route entries accordingly.
Overall it was AGREED that the route server should greatly improve
the stability of routing between Europe and North America.
Questions of scaling - for example the handling of interactions
between two or three route servers - will eventually have to be
answered, but it is sensible to proceed with the pilot route server
as quickly as possible so as to gather experience with it and to
discover how it can be generalized.
Action items:
* The RIPE NCC has been collecting routing information as part of
their network database for some time. This will be sufficient for
the European route server; the main thrust of this development
will come from there.
* A North American route database - or better, a North American
route server - is needed too. Merit has a pilot project in this
area.
* Support is also needed from European providers. Daniel will pass
a RIPE document requesting route information to them next month.
* The prototype route server will be located on the Washington MAE,
which will be used as a testbed for the GIX / Route Server pilot.
A Unix system is needed there to run the route server.
* Involvement in the project by US network providers is needed, so
as to help sort out any North America - Europe interaction
problems as they occur.
* The proposal document
(Bates/Karrenberg/Lothberg/Stockman/Terpstra) will be revised.
Tony Bates will put it into an ftp server, and advertise its
whereabouts to the RIPE mailing list. Elise will copy this
advertisment to the IEPG mailing list.
The meeting adjourned for lunch at 1230.
After lunch the meeting resumed (without the invited guests) at
1408. Those present were:
Guy Almes FARnet almes@ans.net
Tony Bates ULCC t.bates@noc.ulcc.ac.uk
Nevil Brownlee Tuia nevil@aukuni.ac.nz
Brian Carpenter CERN brian@dxcern.cern.ch
Peter Dawe PIPEX peter@pipex.co.uk
Gopi Garge ERnet-India gopi@ece.iisc.ernet.in
Elise Gerich Merit/NSFnet epg@merit.edu
Masayoshi Gohara NACSIS/SINET mg@sinet.ad.jp
Tony Hain ESnet/LLNL hain@es.net
Daniel Karrenberg RIPE NCC karrenberg@ripe.net
Peter Lothberg EBONE NOC roll@stupi.se
Jun Matsukata ISAS, NACSIS/SINET jm@eng.isas.ac.jp
Jun Murai Wide Project/Japan jun@wide.ad.jp
Joo Y. Song HANA jysong@ring.kotel.co.kr
Bernhard Stockman EBONE boss@ebone.net
Fumio Teraoko SonyCSL/Wide tera@csl.sony.co.jp
Marten Terpstra RIPE NCC marten@ripe.net
Claudio Topolcic CNRI/FEPG topolcic@cnri.reston.va.us
2. Presentation on ERnet
---------------------
Gopi Garge (Indian Institute of Science, Bangalore) gave a
presentation on ERnet, India's Education and Research Network. This
began in 1985 with dial-up slow-speed lines and is now running using
higher-speed leased lines, including a 64kbps leased line from
Bombay to Alternet.
Other networks in India are
INET - the public X.25 network
SOFTNET - run by the Department of Electronics; a commercial IP
provider
INDONET - run by the Computer Maintenance Corporation; a closed
network for IBM users
NICNET - a government network run by the National Information
Center
The next phase of ERnet's development is a VSAT network linking
eight sites into a star X.25 network running at 9600 bps. ERnet is
promoting the use of OSI protocols across Asia; to this end they aim
to consolidate Internet and OSI services by 15 Jun 93. Future plans
include a country-wide high-speed backbone for data networking.
Comments from the meeting centered on the use of VSAT for a small IP
network, which doesn't seem to have been done elsewhere. If this
works well it could be a useful model for other countries such as
Eastern Europe, South America. The OSI emphasis is also interesting
since we don't really have a lot of ISO implementation experience.
3. IP Addressing Issues
--------------------
RFC 1366, Guidelines for Management of IP Address Space, was
discussed: no changes were suggested.
RFC 1367, Schedule for IP Address Space Guidelines, was considered.
Claudio explained that the proposed timetable was based on the
realization that address aggregation will be possible by mid-'93.
After some discussion it was AGREED that the timetable is
reasonable. All IEPG members are requested to put pressure on their
router suppliers to deliver reliable BGP4 implementations before 6
June 93.
Daniel presented a short report on the RIPE NCC. RIPE is supported
by all the European IP services providers; it's NCC has a staff of
three in Amsterdam, and has been functioning as an IP Address
registry (including registering route information) for four months
now. Overall the contacts between NCC and the providers has proved
very beneficial.
The NCC have already established local registries with service
providers in order to assign class C network numbers hierarchically
and thus suitable for aggregation. In this respect Europe has
already reached stage 3 of the schedule (RFC1367). This makes it
especially critical for Europe that the CIDR aggregation mechanisms
be available on time.
When allocating Class B addresses they are a lot stricter than the
RFC implies! They have assigned about 50 Class B addresses in the
last four months (including 19 to the Deutsche Bundespost); in each
case the applicants had submitted clear engineering plans
demonstrating their needs. Detailed information about this can be
obtained from the RIPE NCC quarterly reports available on
ftp.ripe.net. The most recent one (document ripe-73) gives a
detailed overview of local registries and network number
assignments.
Daniel commented that it would be helpful to set aside some
addresses for networks which will never be connected. A small
number of these could be used by many networks. Peter Dawe
suggested we should recommend setting aside some Class C addresses
now for this purpose. There was general agreement that this is
useful and Daniel offered to author an RFC about the subject
together with those not afraid to have their name associated with
such an inelegant thing.
Bernhard asked whether he should inform the BGP Deployment Working
Group that the IEPG supports BGP deployment as per RFC 1367. It was
AGREED that he should.
4. Review of the IEPG Workplan
---------------------------
The IEPG workplan was reviewed, and the items in it classified into
two categories: (T) for items which remain as high-priority ones,
and (B) for items which are being worked on by other groups, and
which are therefore low-priority items for the IEPG. The resulting
lists are:
(T) items
---------
1. Infrastructure
1.1 Global routing
1.2 Global DNS (there is a BOF at IETF)
1.3 Global address and route registration
1.4 Operational Impact of IP Transitions <== new item
2. Operations
2.3 Statistics and Measurement
6. Scaling and Forecasting <== new item
(B) items
---------
2. Operations
2.1, 2.2 NOC & NIC Coordination
3. Applications and Services
4. Multiprotocol Integration
5. Enhanced Network Infrastructure
During the workplan discussion two new (T) items were added.
'Operational Impact of IP Transitions' was put forward by Peter
Ford, who felt that the IEPG should generate input for ORAD to make
its members aware of any problems we could see arising from any IP
transition scheme which might be considered.
The second new item, 'Scaling and Forecasting,' was suggested by
Peter Dawe and supported by Daniel. It would be worthwhile getting
the CCIRN to fund someone to survey the network providers, asking
for their predictions as to the number of networks and routes they
have, and expect to have in a year. If done on an annual basis this
would provide valuable input for global network planning.
5. IEPG Current Status and Framework
---------------------------------
Following recent discussions on the iepg mailing list (triggered by
a recent note from Hans-Werner Braun) it was decided to consider
whether changes in the way the IEPG is structured would be
worthwhile. The main topic of concern here was the rapid increase
in the number of 'general,' i.e. non-R&D network providers. It was
felt that by not having general providers represented on the IEPG we
are reducing our ability to support 'ubiquitous connectivity.'
Possible changes could be:
* Create a new ad hoc organization including existing IEPG members
and others.
* Notify the CCIRN that IEPG membership should be widened to
include representatives of general network providers.
After considerable discussion the meeting AGREED to request Elise,
Bernhard and Geoff to write to CCIRN informing them that "it is the
intent of IEPG to increase its membership." It was also AGREED that
the IEPG co-chairs would make the decisions as to which
organizations should be invited.
6. Action List for the GIX Proposal
--------------------------------
The following actions were agreed upon:
* Peter Lothberg, Peter Ford and Guy Almes will finish their GIX
Proposal paper. They will publish this (as an Informational RFC?)
and say that the pilot project is proceeding, and that there are
lots of aspects of the proposal which need further work..
* Tony Bates, Peter Lothberg, Daniel Karrenberg et al. will publish
their paper as a 'rough draft,' and commence work on the route
server.
There being no further business, the meeting finished at about 1750.
Minutes taken by Nevil Brownlee.