Draft minutes 6th EOF meeting

[Havard Eidnes <]

Hi,

please find enclosed the draft minutes from the 6th EOF meeting
held in Rome earlier this month.  Your comments are welcome.

Regards,

- Håvard
							-*- Text -*-

European Operators Forum

Minutes of the 6th EOF meeting				Håvard Eidnes

Held at CNR, Rome, Monday 8 May 1995			NORDUnet/Uninett
							18 May 1995

------------------------------

Atendees:

Christian Panigl	UniVie/ACOnet	panigl@localhost
Wilfried Woeber		UniVie/ACOnet	woeber@localhost
Bernard Tuy		RENATER		tuy@localhost
Marc Pcihon		TRANSPAC	pichon@localhost
Franck Pradal		France Telecom	pradal@localhost
Van Huong Nguyen	TRANSPAC	nguyen@localhost
Bruno Malaval		TRANSPAC	malanval@localhost
Erik-Jan Bos		SURFnet		bos@localhost
Mike Norris		HEAnet		mnorris@localhost
Balazs Martos		HUNGARNET	martos@localhost
Keith Mithcell		PIPEX		keith@localhost
William Tevie		Network Computer Systems     tevie@localhost
Pekka Isomaki		Telecom Finland	Pekka.Isomaki@localhost
Jarmo Oksanen		Telecom Finland	Jarmo.Oksanen@localhost
Gianfranco Pesce	Aqora' Telematica	g.pesce@localhost
Leonid Yegostin		RELCOM/IHEP	egoshin@localhost
Dmitry Avdeyev		RADIO-MSU	ada@localhost
Janos Zsako		BankNet		zsako@localhost
Graca Carvalho		FCCN/RCCN	graca@localhost
Damir Pobric		CPR-Pisa,Haly	pobric@localhost
Nigel Titley		BTnet		Nigel.Titley@localhost
Oliver Smith		Demon		oliver@localhost
Mandy Antoinette	Demon Internet	mandy@localhost
Cliff Stanford		Demon		cliff@localhost
Thomas Stalder		SWITCH		stalder@localhost
Willi Huber		SWITCH		huber@localhost
Ruediger Volk		Deutsche Telecom	rv@localhost
Stephan Biesbroeck	BELNET		stephan@localhost
Willem van der Scheun	SARA		scheun@localhost
Michael Behringer	DANTE		M.H.Behringer@localhost
Bill Cessna		IBM Global Network	wcessna@localhost
Brian White		IGM Global Network	brianw@localhost
Jiri Orsag		EUnet/CZ	ors@localhost
Petr Mosnicka		EUnet/CZ	moshna@localhost
Klaus Landefeld		Nacamar		landi@localhost
Ngamabadi Kia Ramin	ULB		najmabadi@localhost
Lars-Johan Liman	EBONE NOC	liman@localhost
Havard Eidnes		NORDUnet/Uninett Havard.Eidnes@localhost
Anne Lord		RIPE NCC	anne@localhost
Mirjam Kuhne		RIPE NCC	mir@localhost
Benoit Grange		NIC FR		Benoit.Grange@localhost
Philippe Lubrano	NIC FR		Philippe.Lubrano@localhost
Francis Dupont		INRIA		Francis.Dupont@localhost
Damiusz Piotrowski	NASK		davek@localhost
Ugo Manfredi		DS Net		um@localhost
Giuseppe Miotti		INTESA		gmiotti@localhost
Marco Poet		INTESA		mpoet@localhost
Felipe Garcia		SERVICOM	Felipe.Garcia@localhost
Gustavo Sanchez		SERVICOM	gustavo@localhost
Geert Jan de Groot	RIPE NCC	GeertJan.deGroot@localhost
Stefan Lacher		EUNETCOM	slacher@localhost
Pietro Marchionni	SELFIN		info@localhost
Tibor Weis		SANET		tibor@localhost
Jan Haerung		CESNET		hh@localhost
Andrey Lazaryanz	Yaroslavl University	lazar@localhost
Valentina Morozova	Yaroslavl University	frost@localhost
Dimitri Sidelnikov	FREEnet		sid@localhost
Bettina Kauth		DFN-NOC		kauth@localhost
Sabina Dolderer		DE-NIC		sabina.dolderer@localhost
Juergen Rauschenbach	DFN		jran@localhost
Nevenko Bartolincic	CARNet		Nevenko.Bartolincic@localhost
Ivan Maric		CARNet		ivan.maric@localhost
Ruzica Vukovic		CARNet		Ruzica.Vukovic@localhost
Jean-Michel Juanigot	CERN		jimi@localhost
Peder Chr. Norgaard	Telebit comp. A/S	pcn@localhost
Andrea Borgato		I.Net S.p.A	andrea@localhost
Giuliana Tamorri	GARR-NIS	tamorri@localhost
Maurizio Martinelli	GARR-NIS	martinelli@localhost
Edoardo Calia		Politecnico Torino	calia@localhost
Valeria Rossi		CILEA		valeria@localhost
Giovanni Armanio	GARR-NIS	armannio@localhost
Daniele Vannozzi	GARR-NIS	vannozzi@localhost
Carmelo Notaro		Ist.sup Sanita	notcarm@localhost
Fabrizio Pia		CSI-Piemonte	Fabrizio.Pia@localhost

------------------------------


 - Peter Lötberg (PL) welcomed the participants

 - Håvard Eidnes (HE) was appointed minute-taker

 - Previous minutes

   HE commented that he had received a minor correction to the
   minutes, which will be included in the final version.  Given that
   this is a minor change, the previous minutes as included in the
   RIPE-20 minutes were approved.

 - PL informed about the next EOF meeting, which will be held in
   Stockholm just before the next IEPG meeting (which is again just
   before the IETF meeting).  The dates are:

   EOF meeting,  Friday 14 July, 0930 at KTH, Stockholm
   IEPG meeting, Sunday 16 July, 0930 at KTH, Stockholm

   All the EOF meeting participants were invited to participate in the
   IEPG meeting.

 - NSFnet shutdown

   Elise Gerich gave a short presentation of "the sequence of events
   as viewed from Merit".  After a meeting at the NSF the shutdown of
   the peering sessions at the NSFnet attachment points were advanced
   a week, so that a fallback could be implemented for a (single) week
   if any problems should arise.  NCAR asked for an extension (which
   was granted), and the ENSS at MAE-East had problems running with
   the modified configuration, so the turnoff was pushed back from
   friday to tuesday.  At tuesday when the peering sessions were
   turned off, a single AS (AS 22) was "lost", connectivity was
   restored for this AS some 48 hours later.  Otherwise there were
   some single-network problems which were dealt with on a case by
   case basis.  Thus, at the 30th of April, a quiet "non-event"
   happened when the NSFnet routers were turned off.

   PL then proceeded to give an overview how ICM was affected by
   the change.  Before the NSFnet shutdown, traffic from ICM to
   FIX-East averaged 20-25 Mbit/s on average, much of it going to
   ANS.  Since ANS would not be present at FIX-East after the
   NSFnet service shutdown and ICM was not present at MAE-East+,
   severe overloading of SprintLink's connection to MAE-East+ was
   expected if nothing was done.  Thus, a new T3 connection from
   ICM to MAE-East+ was installed, and since the "one week push
   forward" was done this had to be installed on very short
   notice.  Luckily Sprint managed to install the circuit in
   time.  Later a separate T3 has been added between ICM and the
   NY NAP at Pennsauken.

   PL concluded that the lesson to be taken from the shutdown of the
   NSFnet service is that turning off a sponsored backbone service and
   letting the commercial service providers compete on a level playing
   field actually works quite well and has benefits to the users.

   PL also explained the plans for implementing a "Euro-NAP" at the
   Pennsauken, NY Sprint facility.  The Pennsauken installation has
   better survivability than the MFS facilities in Wasthington DC, and
   prices and rack space are both readily available.  Due to this
   facility also housing a US NAP in the post-NSFnet architecture it
   is a well-suited place to terminate high-speed european circuits.

 - European GIX deployment

   PL claimed that a structure similar to the new US architecture
   would be highly beneficial for the development of european Internet
   intfrastructure.  He presented a list of proposed european *IX
   point locations:

	- Pennsauken, NY, US
	- Telehouse, London, UK
	- KTH/Tele2, Stockholm, SE
	? France Telecom / Metropolitan Fiber Systems, Paris, FR
	? British Telecom, Amsterdam, NL

   with the two last ones as more uncertain than the other three.  In
   such an architecture, major service providers need to "touch down"
   on all the exchange points and can there establish peer
   arrangements.  Service providers not wishing to connect to multiple
   exchange points typically have to subscribe to some kind of transit
   service to reach where they want to go.

   Also, putting major information servers at or near such *IX points
   may be an important aspect of optimizing the network usage within
   Europe.

   The ground rules for such an architecture and the specific
   locations of the *IX points needs to be written down and agreed to
   by the major players, and this should be done ASAP.  PL asked for
   volunteers at this point, noone came forward during the meeting,
   though.  PL mentioned that there would likely be a meeting in
   Amsterdam the 6th of June where the providers likely to pull lines
   to routers placed on the *IX points would be invited.

 - Service provider interfaces

   PL presented a categorization of service provider interfaces:

   Type 1: statically routed from the service provider or lives within
	the service provider's IGP, thus in the same AS as the service
	provider.

   Type 2: single-homed customer using BGP toward the service provider
	via CPE belonging to service provider but in customer's AS.

   Type 3: multi-homed customer, CPE in service provider's own AS,
	local external BGP peering at customer premises.

   Most customers will be in category 1, relatively few will be in
   category 3.  Category 2 is usually not recommended, as it "burns"
   an AS number.  PL commented that that can be fixed by hiding the
   existence of the AS externally, but this configuration permits to
   solve a problem where BGP is the appropriate tool.

   Also, a definition of where the access point for the service is or
   can be in each case was presented.  The slides from the
   presentation are available at

	http://www.stupi.se/

   under "Peter's opinions", and can be freely copied and used.

 - Swedish affairs

   PL presented some measurement of the performance going to customers
   of the swedish operators (connected at 64 kbit/s).  The graphs
   showed the result of tallying ping RTTs over 24 hours.  PL
   commented that one particular service provider's rather large
   variation in RTTs would make TCP behave sub-optimally, and that the
   likely reason would be that the service provider uses Frame Relay
   for delivering the IP service.  In PL's opinion that didn't make
   too much sense.  That particular service provider's packet loss
   rate was also in the 10% range, whereas the others typically were
   in the 1% range.  When the experiment was repeated from the US, the
   packet loss rates were in the 15-25% range (ugh!).

   PL also showed a graph displaying the growth of the number of new
   registered domains under .SE -- during the two first months in 1995
   there was almost registered as many new domains as in the entire
   1994.

 - Any Other Buisness

   Net 39 experiment
   -----------------
   Current predictions are that we will run out of IPv4 addresses by
   2018 plus-minus 8 years.  This however assumes that the remaining
   class A address space can be subnetted and doled out to different
   service providers for assignment to their customers.  In order to
   find and fix any problems in doing this the class A network number
   39 has been temporarily assigned for this testing.  Every owner of
   an AS number has the subnet 39.<AS & 0x7fff>.0/24.  The assignment
   is as mentioned temporary, and lasts until 951131.  The point is to
   experiment without directly involving customers.  Currently gated
   and Cisco code is being tested, Wellfleet was mentioned as a
   candidate for testing as well.

   Keith Mitchell from PIPEX commented that PIPEX has been announcing
   a subnet of the class A network 9.0.0.0 for some time already (in
   "production mode"), and the only problem so far was a slight delay
   in processing of the NACR for the NSFnet service.

   The message that multihomed EGP and BGP-3 users will lose out big
   on this one needs to come out *now*.

   There has been set up a mailing list to coordinate and discuss this
   experiment: exp39@localhost.  Send to exp39-request@localhost to join.
   A hint to Cisco users: do "ip classless" before participating in
   the experiment.  Some hosts already on net 39 were mentioned:

	anka.stupi.se
	exp39.ripe.net
	postel.wind.surfnet.nl


   Exchange point meeting
   ----------------------
   As mentioned above, PL is going to call a meeting for discussing
   the exchange point architecture and placement for europe on the 6th
   of June in Amsterdam.