Reqs: Virtual Routing Domains (overlapping address space)

[Jim.Bound@nokia.com]

Folks,

Wondering if something I have seen looking at next gen router  AD
development with IPv6 affects a requirement and possibly another issue for
multihoming.

Scenario:

ISP1                                 ISP2
Core


           Aggregagte Router (AR)
Border


Phone - Multidwelling Unit - PDA - Roaming Lap Top               End User

This is an oversimplification scenario but hopefully will be useful to
convey the issue or req.

ISP1 and ISP2 are at the Core of the network in this scenario
AR is at the Border of that Core
Nodes are at the End User level to the Border

Reality today is that many ARs today have to deal with providing services to
the Core and in most cases its not because the End User is multihomed but
simply because traffic is being sent to multiple ISPs because of where the
function for the user lives on the network beyond the Core.  I think we can
assume that it is not a property to consider if the End User is wireline or
wireless other than wireless requires good performance solutions
(performance being many parts to work),

There are two cases to the scenario

  A.  ISP1 and ISP2 do not share traffic and are independent to each other.
  B.  ISP1 and ISP2 do share traffic and agree to forward the others
traffic. 

The issue is Case A for this issue as I believe we already are aware and
working on Case B though this issue may add value to that discussion.

>From AR to ISP1 and ISP2 tunnels are used like L2TP or MPLS. I won't get
into the reasons for this and I would be glad if we could optimize this in
the router community with IPv6 as its a pain (not MPLS but the tunnel
software).  One reason is ISP1 and ISP2 may give End Users today the same
IPv4 Private Address space so AR has to deal with another level of
indirection at AR to determine forwarding the packet to the Core this is
where the virtual routing domain issue comes in to play.  Hopefully End
Users will use IPv6 Global Addresses and that reason for tunnels goes away.
But there are other reasons some believe why the tunnels are useful (e.g.
one virtual pipeline to specific Core routers).

Can we make a suggested or noted requirement that the Border not permit
specific host routes be propogagted to and from the Core?

Do we need optimization requirement so the Core ISPs can redirect traffic to
other provider with specific multihoming characteristics to the Border
routers?

regards,
/jim