RIPE 42

Please mail comments/suggestions on:

• content to the Chair of the working group.
• format to [email protected].

RIPE 41 / EIX WG / Grand Ballroom / Wed 16/01/2002

Chairman: Fearghas MacKay
Scribe: David Knight <[email protected]>

Presentations are available at:
http://www.ripe.net/ripe/meetings/archive/ripe-42/presentations/index.html#eix

- Agenda bashing

- Bill Manning / Peering Contacts DB

Card swap parties ..
Noticeable dirth of European participants
Bill will collect contacts to have Europeans added to the DB

- CO Chairs

Fearghas Proposes: Mike Hughes / Christian Panigl

Approved



- BCP update - Nic Lewis

Has been discused last couple of meetings
Large amount of work
Proving diffcult to actually find best current practices

- Progress

There is lots of overlap with EuroIX
A dialog is maintained with EuroIX

Outcome - Nic has been asked to assist making a website with them,
and will help to harmonise application procedures

- Breakdown

Working to harmonise joining procedures
Standardise tech procedures

- Considerations

Concerns about adoption
EuroIX is paid for, results are more quickly forthcoming

Nic thinks EuroIX is a good vehicle for this work it may serve the
community better than the bcp, he sugests transferring the work to
EuroIX

Nic is not directly employed by EuroIX
The work remains visible to the community

- No objections

- Questions ?

None



- IPv6 Allocation Policy for IXPs - Fearghas MacKay

- Fearghas asks if there is any feedback

None



- Tools Wish List - Mike Hughes

Version 2 of the wishlist is up on the WG site

Keith has promised input, as has Christian, more to come from the
meeting with Cisco

Cisco have started a list, poor account mgmt meant this didn't get to
the right parts of cisco , now it has they are interested in this info

People from amsix / milan / exchange poiunt and arnold from decix joined
in meeting with cisco

Cisco realise that they neglected the community of late, due likely to
their poor market share in euro IX's

It was a very good event ...

Cisco alluded to having more events ...

Something good has come from the document already !

- Christian -

There will be a mailing list of exchange operators

- Mike -

A number of IXP operators were invited but couldn't make it
so close to ripe41

Cisco alluded to Amsterdam as next location

It's a good sign that vendors are prepared to work with IXPs

Extreme and Foundry seem to be taking things from the list and
implementing

Mike is the list editor, contact him to ad to the list



- IXP Presentations

Presentations are available at:
http://www.ripe.net/ripe/meetings/archive/ripe-42/presentations/index.html#eix

- BLNX - First Presentation - Stefan

It is publically founded limited company (gmbh)
The majority of shares held by founders

BLNX is a classical IX
They are neutral
They are independent of local carriers

- Ports

100/1000 IEEE802.3
E3/T3/STM-1 SDH

- Free Stuff

Stratum 1 clock
Ticketised support

- Where

Berlin

4 locations

Friedrichstrasse east
Gradestrasse south
Lutzowstrasse west
Alboinstrasse west

They are looking for wireless solutions as connections are expensive
and difficult

- They have

good business plan
need for backup
growing bandwidth
location near new customers / coming east germany and poland

Talking to people in Poland who are interested in coming to Berlin


- Contracts

IPBerlin, Akamai, go.ana
Signed LOI's
Four big ISP / 3 midsize ISP

- The Truth

They want to get rich
It can be done
Berlin is the growing centre as you extend to the east

www.blnx.net


- Questions

4 locations for a startup seems like a lot, sounds like some colo
arrangements are in place, are the sites managed by other parties,
what is the relationship ? Are you neural from the colo ?

- Nobody is here now, no 24/7 support
they will rent space and install own techies
carriers here now: level3, computel

If you are considering wireless .. does this suggest
something wrong with your business model ? Does that
seem reasonable to you ?

- There is some concern, that they are not reliable.
There is a problem in berlin, carriers put in much fibre,
but it is expensive.

If you cannot reliably interconnect 4 locations, should you
really be at 4 locations ? I have not heard of wireless used
in this way before.

- This is a proven technology, microwave works, it is reliable to
STM-1 but higher bandwidth is the limitation ...

- At locations 1/2 must be links lower than stm1m, once we
have money we can get fibre.

As a point of clarification, when you say wireless some people may
assume you mean 802.11b, you mean traditional telco microwave ?

- Yes, and we do this with xtreme / cisco


- FICIX - Jorme Mellin Chair - [email protected]

- History

Founded in 1993 by Telecom Finland / Helsinki Telephone co
10 mbps ether switch
registry for .fi ccTLD

1996 Upgraded to ATM

1997 .fi moved to FICORA

1999 Aditional IX founded / Cisco 8540, 155ATM and 622ATM

2001 Association founded / 11 operators signed

- Status

4 members
Carrier independent
155atm and 622atm available FICIX1 / FICIX2
IPv4 and IPv6 matrix in palce
KONQwuest is service provider of physical maintenance
Non profit association
Elected trustees

- Traffic stats

http://stats1.kpnqwuest.fi/

- Joining

Apply to [email protected]
Must be present at both peering points

- 2 member classes

Full no peering agreements needed

Supporting Member / Peering agreements needed

- Details

No colo space nor fibre

Fibre available from several proveders to both locations

Expected port cost ifs 500 Euro after FICIX1 upgrade to gigE

ATM155 avail til q4/02

- Future

IPv6 in place ... peering exists

Thinking about multicast

Possible relocation of ficix1 ...

Could also provide colo space or move to area with plenty colo space

Hoping for somewhere in the capital area

- URL

www.ficix.fi

- Questions

None


- AMSIX - Henk Steenman

- Status

Not much infrastructure change since RIPE40

4 sites: nikhef / sara / telectuy / global switch

Nikhef - Sara quad gE connect, other sites double gE

- Stats

124 members
177 routers

Most ports at nihef (90) / sara (66)

Yesterday for the frst time they exceede 6gig/s

They are growing fast

As a consequence, they as looking to upggrade the connects to the 4
sites

They have a seperate vlan for gprs roaming exchange, currenty 18
ports

Managing private .gprs root server

Have been assigned a /84 for ipv6

AS1200 moved to dual stack, all existing ipv6 peerings will move
to the production router

- Main changes are organisational ...

Operational mgmt was always outsourced, now they have their own noc
they manage the exchange 24/7

AMSIX office is now located at Westeinde 12 in central Amsterdam
which houses the office, noc and euroix

Total staff is now 9

- Questions ?

None


- DECIX - Arnold

- As last time ...

They have been working for a longtime on a new global framework

They have scheduled meetings for mid February

Decix is going now and they need a good legal framework

Membership is stable, some left due to mergers, etc

They hope to hit 100 mark in the middle of this year

- Operational

They must replace their route server

Must have full control over prefixes

Second router server is under development, will be installed for
increased resilience, as well as new features, not just for
filtering but also prefix filtering, they replace the cisco
filter with a new siemens / spacrc box

End last year, 30 Dec they were hit bya mutlicast storm. It took
them 12 hours to isolate and control. It's not yet understood what
happened, they could identify all involved ports were on gsr and gig
ether, there is curently a cisco case open, they hope to learn more,
they will upgrade to latest ios to support mcast suppression, since 6.5
this is possible. This demonstrated the lack of tools to identify
these problems, they already took the initiative and said to euroix
that they need the tools.

- Stats

Traffic is increasing, but not much since last time

Peaking at almost 5 gigs

- Projects

MCAST started mid last year, ten members involved
Trial will continue until mid this year, then enter prod as a
regular service

IPV6 started last October, no participants, the project will
continue until the end of this year

Surprised by the lack of interest

- Questions ?

None


- CIXP - Paolo

- Since last time ...

Much industry consolidation means router size has increased

First non cisco in production during 2001

First gigE ports allocated

Growth in WDM kit used ty telco operators

Number of members is constant

Steady growth in traffic

- Status

They are partnered with telehouse switzerland, expansion site in
downtown geneva, no gigE there yet

They plan to setup vlans for backtoback peerings

More gig capacity in 2002, only price difference is the install
charge

They participate in EuroIX

They cooperate with other euro research projects

Also some interest from commercial in IPv6

They plan to have ipv6 research peerings

New URL http://www.cixp.ch/ - redesign in cooperation with
telehouse is planned

Looking glass is operational, only for members, peering with
AS513

Renewed interest in the mcast testbed, a real service durin 2002

- Questions ?

None


- INXS - Udo <[email protected]>

- History and overview

One of the oldest exchanges in Germany

Currently 42 members

New members wating to connect

Open house policy for any carrier
All circuits terminate in a seperate room
Patching done by C&W staff

ISPs may make INXS their POP in Munich

Memebers may buy and sell traffic / may have private peerings so
long as they don't sell upstrweam and INXS are switching fabric

Member must have their own ASN, must be a LIR, must house a router
at INXS, switches or LAN not allowed

No obligation to peer

No route server yet

Looking to participation in RIPE RIS

- Fees

Unchanged during euro change ...

- The exchange

Cisco Catalyst 6509
One ip per customer
10/100/1000 ports

- Website

http://www.inxs.de

Members area
Operators area
Mailing list archives

- MCAST

MCAST trial only regular INXS members
No fees during test phase
Limited to 1/3 of INXS members
Only 10/100 ports
Test prolonged to mid 2002
Dedicated vlan
In operation since oct 2000

- IPv6

Same rules as mcast
Additional rack for ipv6 routers
In operation since sep 2000
5 isps connected
Currently using C&W ipv6 allocation

- Status

Growth in members increasing steadily

Some members lost to mergers

Some new members though

- Contact

Udo <[email protected]>

- Announcements

INX-HH in Hamburg is INXS Hamburg

Locations not connected, no plans to do so

Same contract and service will be delivered as in Munich

More details on the web ...


- Questions ?

None


- LINX - Mike Hughes <[email protected]>

- Since last time ...

New London Office

Some kit going to a 9th location

10 gig is arriving

Some work done to improve resiliency

Passive private interconnects

Inter provider problem resolution being worked on, Rob suggested
feeding this info to the group

Having problems getting into new ofice

Main place for engineers / meetings / it's easily reachable

No changes to phone numbers

New Redbus location at Sovereign House London
50,00 square foot facility, will handle overflow from the original
redbus facility which is full

Diverse connectivity

Full size switches from

Got first 10gig cards from Foundry, tested at HQ in December

Tested at Foundry UK on Jan 4th

Main core can be upgraded to 10gig throughout

They operate dual vendor core

Currently deploying 8gig CWDM solution as it is available NOW

Xenpac will bring down the price of the optics and ergo the cost
for 10gig cards

Currently kit comes from extreme and foundry

They are considering a connection diversion policy

Everyone which is single homed will be in one lan with everyone else

The idea being better vendor support

The main archtecture of the new split network is two 8/10gig cores
with diverse fibre routes

- New service

Passive Private Interconnect, layer1 managed fibre

This solves a problem with adhoc private interconnects, this
location is londons oldest colo, risers and voids are full of
cables, this helps the members out

As telehouse don't provide this service linx will provide another
private interconnect here too, making use oif existing fibre bundles
in each building, interconnection by fibre supplied by telehouse
linx will pass on the extra rental to the member, advantage being
that as a private fibre the member may run whatever thyt like over
it

- Traffic

Peak traffic this morning is just over 13gig/sec

Has grown steadily since July, Code Red ? ;)

This illustrates why 10gig is needed

Memebership is 125

Some people changeing their models, losing members

Net gain in 2001 though

Good interest in LINX meetings

At last meeting ran a peering sim game

- Tech Challenges

A ethernet switch was connected which shouldnt hhave been there
One switch reacted very badly, caused ripple effects, took 8 hours
for things to settle

This raises questions about how to do these interconnects safely


- Ongoing development

- MCAST

Still a seperate vlan
Can extend over vlan when needed
Chicken and egg scenario

- IPV6

Space for peeroing allocated
Chris will do some experiments

- Questions ?

Interest in pasive private ?

- LINX does this free for thiscurrent billing period, no charge
if you join now


- NDIX - Rudolf van der Berg <[email protected]>

-

Based in Enschede - current site - will extend to munster and start
a second site will be done with the de local govt to ensure neutrality

ndix is neutral - startext by the local govt

Location is in telehaus in munster, neutral location

Tech matters resolved

Hardest thing was a line from enschede to munster, now they have
fibre

Up to the politicians to getr rid o the red tape

Operational by the end of the second quarter 2002

Interesing new dev in esnchede is wireless testbed, funded by govt
started by University of Twente, ndi will provde head end on a
seperate vlan parties may connect test kit and connect through
to the internet

Not just about the technology but about the application too,
hospitals are involved in the tests

Versatel is connected, two more telcos this quarter

Rollout of wireless testbed in q1 2001

The IX is running well with no downtime so far

- Questions

Was there a special reason to go to munster ?

- Enschede is 5km from the German border
Germany seemed a good idea
They got in touch with Germans, they were interested, Munster
evolved from this

I know the area, I don't think that munster is attractive ?

- They disagree

Munster already hs 4 telcos, several wireless operators present in
the area, Universities at either end cooperating
Only 20km from dortmund where stuff is happening
Close to the Rhurgebied ..

Why Munster and not Dusseldorf ?

- Cost effective


- NIX.CZ - Josef Chomyn

- Status

Located in Prague
Non profit
Est August 1996
28 members
2 left Dec 2001
2 locations

10/100/1000 ports
Using Cisco Catalysts, 2948/6006/6006/4003
Gig interconnect, Cat 6000's at either end
100 interconnect, using old switches

- Currently preparing

gig for customers
contract for 1st employee - Managing Director
workin on draft - nix.cz concept for the future

- Contact

www.nix.cz
[email protected]


- TIX

- Status

Not that much news

Got a bigiron switch

3 gig connections
40 fast ether

No ipv6, have not requested addresses

Run a .ch secondary

- Questions ?

None


- VIX - Christian Panigl

- Status

50 people at workshop last year

Found out som interest in remote connects to the VIX, they are
looking at new technologies

Looking at connect agreements to allow remote router connects

- Questions

None


- Exchange Point - Keith Mitchell

- Brief update

Network up for 8 months
Free trial over
17 customers, making money now
Hitting 100meg level
Met SLA targets

- Sites

London: Redbus / Telehouse East

- Traffic

peak 93.53 MB/s
avg ~60 MB/s


Restatemnt of neutrality principles

Lists types of interconnect

Finding place in the market for virtual private interconnect, could
be used to make a virtual private exchange group

Lists service offerings

Describes VLAN based services, see demand for p2p vpi

Value add, cisco route collector

- MCAST

We won't get thisto work properly until the vendors get together on
this

Community needs to bang vendor heads

Extreme is not getting it's act togeter

Keith desn't care about the differeing technologies, just wants
something that works

- New services

MetroXP Multipeer - optional mip router

4 participants

- IPv6

Auto adress allocations to all customers
Would like to get a collector router running

Hosting sec ns for .uk

- Customer private web pages

http://customer.xchangepoint.net
Dynamic customer view
Who offers transit

Customers happy tpo shre data with other customers not necesarily
with everyone

They will host a seminar in London on Feb 7th

- Questions ?

None



- Ethernet over Anything - Peter - Ebone - <[email protected]>

- 2 questions for IXPs ...

- You can set up p2p ethernet from anywhere to anywhere ...

For IX's you could onnect form anywere in the world, you don't
need the router on the IX itself, Christian mentioned he will alllow
remote connections in future

- Not necessarily over anything

Virtual routers, we will see it on IXs in future, including non ether
based IXs, this ould arrive soon, whats the comments ?

- Fearghas - Maybe people should think about it and comment on the
list?

As long as this doesn't harm the IX then why not

- It coul harm the IX in that with the lack of kit necesary it
might harm IX kit sales, an IX doesnt need to be a physical
placem it could be virtual, anywhere, conecing anything

Whats the opinion ?

- The possible harm was meant technically, not business models

- Mike - A couple of comments from the Cisco day

Link Layer is not transmitted to the IX switch

What we need is OAM type thing to pass through to allow
dropping at both ends if there is a failure in the middle

LINX is responsible for cable fault etc, dealing with this
virtually really needs link failure to work, chasing ghosts
could consume a lot of man hours

Managing people perception of this, remote routers cause
unexpected latency they don't see that there is a
extranational hop

- Bill Maning - In other disciplines, how to reduce the apparent latency ?

Others have done interesting things, but the only
way involves trying to lie, these solutions appear
attractive, but they always seem to fail

Moving ether frames, what happens when you pass 10gigs
this will cause the saem ATM sucks opinions ...


- EuroIX - John

- Status

Non profit
Established Jun 2001
Provide a portal for euro ixps
20 ixp members

725 listed isps ovwer the 20 ixps

Strugled with existing policies

Approach is join now and seek a fee waiver for new ixps

Operational defined as at least 3 peers and a published customer list

The web site is up andis a sucess so far, fair amount of content,
recording a lot of hits ... picked up a lot since the praha meeting
memebers linking to it directly from their own stes ... 32000 hits
since March ... most coming from Euro countries ... a significant
number from US though

More info soon to be added to the IXP section

Conditional approval given to 3 projects

ISP quality of service metrics
IXP monitoring systems
IXP membership application procedure

Not actually begun this yet

Metrics could be contraversial, going to explore though

Monitoring systems would be good

Applications stuff make it easy for new members to join


- Questions ?

None


- Peering Point Interconnection - Keith Mitchell - eXchange Point

- Some ideas for interconnection between ethernet MANs

Not necesarrily a solution for virtual router problems ...

The following is a summary of keiths thoughts

In Europe most IPP based on swithed ethernet, this technology can
be used for other things, colos do it for customer interconnects,
not necessarily peers

We have competing ISPs connecting, do we want competing IXPs
connecting?

When is it a good idea to connect these things ?

Traditional wisdom says it's always a stupid idea

We have a lot of colos in europe

For many IPPs its not an option to go everywere, it's convenient
to use this to provide coverage

Why not interconnect ?

It's uncommon in Europe
Only one example in Europe

It's common in US, at MAE east a distributed bunch of colos and
IXPs connect their switches together

Layer2 switch technlogy is not a good thing to do interdomain
routing with.

You only have spanning tree - it sucks

Shared media accidents are very easy and very troublesome

It's very difficult to operate a multi admin single network

What kind of applications would you want this for ?

Alow customers of on IXP peer with those of another

Allow them to buy/sell transit

You could also bridge the exchanges with a router, this is
normally an ISPs job, as an IXP you can avoid conflict of
interest with a multi lateral peering router

People have spoken about:

You have a multi view route server, you could provide transit
between the two

There are lots of posibilities ... some scarier than others

If you interconnect to layer2 switched nets, network hygeine,
max, filters minimum scary things .. block all broadcast traffic
bar arp.

Keith would like switch vendors to allow filtering arp packets
based on ip address

Best tool is vlans ... provides most protection needed

All untagged traffic blocked - dont leak, only explicitly added
to vlan people should be able to talk across the interconnect

STP causes a lot of problems, no filtering or policy constraints,
all changes or problems proagate

Using the interconnect - no stp - maybe use trunking instead

New ieee stp standards 802.1s and 802.1w may help, cisco talkng
about tag stacking, vlans within vlans

IP addressing is not such a hard issue, you dont want to route
between them, if you extend the vlan across the connection you may
as well take your adress range with it, sub interfaces with the
remote address range could work here

It's hard for IXPs to stop customers plugging their kit in

We have a model in the shape of 'peering'

Don't allow arbitrary connections, sit down and hash out a
bilateral agreement ... we need new language here or we get
confused very quickly

Working now with Packet Exchange ... and other parties

White Paper available soon

Keith thinks there is mileage in this ... as a community we should
think about this ...

- Questions ?

- About a route server between nets, it's been done ... the problem
is that the router must forward at line rate at gigE while route
serving or have two routers ... one forwarding ... one serving,
it's complex and expensive ... you could also not have a route
server ... let people statically configure their connections,
ebgp multi-hop

They also tried running bidirectional static map across this router
This is a gross hack .. but it works

- No money is changing hands ... who is paying for the connection ?

This is like a peering agreement ... if you work on the basis of
MetroEther with abundant bandwidth ... the costs are done on a
non settlement basis ...

- Someone has to pay for the fibre ?

Nobody is talking about extranational connects .. they are talking
Metro area at most .. it's easy to justify interconnect between two
admins rather than two sites under one admin


- MAE Services and facilites - Thomas Bechly - WorldCom

List service goals ...

Have around 50 customers in Europe

Using CATS AND GIGE

In the US runnign atm service -- esablished pvcs for direct peering
guaranteed and best effort service ... wth policing

MAE East / West both distrbuted 4 sites / 2 sites

Benefits allow to keep up to capacity

Lists service level benefts ... rules

New Service ... Customer connects Frame encap technology
best pvcs between customers .. moving up the food chain with this
service ... new site opening in new york and chicago

In their model they use atm ... which has issues ... a frame
interface fits the model and provides a clena handoff with savings in
efficiency .. also keeps the exhange platform away from routing
issues ... etc

New service at MAE east this year

More info at www.mae.net ...

- Questions ?

None


- AOB

We will try to have a session with Michael Beringer delivering


Presentations are available at:
http://www.ripe.net/ripe/meetings/archive/ripe-42/presentations/index.html#eix