[ppml] 2005-1:Multi-national Business Enablement

Owen

I think Fred is describing reality for those of us that have global
networks.  Maybe you have a point on the /64's; that could be an option.

But sorry, no global corporation is going to accept a single ISP
solution as currently dictated; not one of us would survive trying to
explain this to our Boards.  That is reality whether the routing works
or not!!  And it is the crux of whether v6 can advance or not.

Take care
Terry

-----Original Message-----
From: Owen DeLong [mailto:owen at delong.com] 
Sent: Saturday, April 23, 2005 12:29 AM
To: Jeroen Massar; Wettling, Fred
Cc: ppml at arin.net
Subject: Re: [ppml] 2005-1:Multi-national Business Enablement

> Let's make a nice normal typical example of a 'multi-national
business':
> 
> Thus there is a company lets name it Example Corp.
> This company has offices (read: sites) all around the world (New York,
> Amsterdam, Paris, London, Tokyo, Canberra, Seoul, Lima, etc). Every
site
> has their own admins so they want a /48 per site, just like every
> enduser with a dsl line, cellphone, or whatever connectivity method
gets
> a /48. As this company is large it also has a lot of employees, and
> these like to dial in to the company network using VPN's. Thus
everytime
> a employee connects, this employees network wants to get connected to
> the company network and thus the VPN gets a /48 routed over it too.
> 
Um... generally, the company should be giving /64s to the employees,
VPNs,
etc., not /48s.  Every end user with a DSL line, generally, should also
be
getting a /64 unless they have need of multiple networks, in which case,
a /48 would be justified.

> Effectively this company will thus need a /32 or similar large sized
> block, just like Google and Microsoft amongst others already have.
> 
Not necessarily, however, this example is _NOT_ the example that 2005-1
is targeted for.  This example could be an LIR.  Now, if the company
wants
to treat each site as a separate ORG, then, those sites might,
individually
be eligible for /48s under 2005-1.

> Now a fun part. The site in Lima doesn't have that much connectivity,
it
> has only a 2mbit SAT uplink. The site in Paris is also not very well
> connected, only a 10mbit leased line.
> 
> The webservers need a 1Gbit connection, because a lot of French people
> are connecting to it etc. Those webservers are located in New York.
> 
> Now where are you going to do your BGP announcements?
> 
> Do remind that the company gets a single /32 and are not supposed to
be
> announcing multiple /48's out of that, as that will break the whole
idea
> of aggregation. Also keep in mind that if you only announce it in New
> York that traffic from the employees summer house in Nice will flow
over
> New York to Paris, introducing a nice 160ms latency for his SSH
> connection. If you announce it in Paris, without limiting it to the
> peers, because then you introduce the latency again, then a lot of
> french people and surrounding areas will go over that teeny 10mbit
> leased line, while they all might want to download that super cool new
> product advertisement movie, which does fit over the 1Gbit pipe at the
> webservers but does not fit over the 10mbit leased line...
> 
If you're going to be an LIR, it comes with the responsibility for
building a backbone sufficient to meet your Intradomain connectivity
needs.  If your dealing with multiple organizations that are diversly
connected, then, topologically they are many small organizations,
not one large one.

Owen


-- 
If it wasn't crypto-signed, it probably didn't come from me.