Hi Robin,

 
> > Adding ISP uplinks to single-homed edge networks can be very
> beneficial.
> 
> But it requires some kind of scheme to make them useful.  Currently,
> the only way to multihome like this is to get PI space and advertise
> it into the DFZ through one ISP or the other.  As more and more
> end-user networks do this, so we have the routing scaling problem.
 
Multi-homing with PA is possible. See RFC3704 section 4 on suggested
solutions. My BRDP Based Routing proposal is an enhancement, it eliminates
careful planning and configuration and the need for tunnels.

 
> > Taking an existing ISP uplink out of service would need more
> attention, but
> > don't say this is impossible or not cost-effective. Let's say: it
> depends.
> > More important: it can be decided site by site.
> 
> I haven't been able to understand this clearly, or why you think it a
> host-based routing scaling solution could pass either set of
> critiques I (and others) have made:
> 
> 1 - It would be impossible to deploy it widely enough (in any
>     reasonable time frame) that there were such a proportion of
>     all hosts (such as 95%, 99% or whatever) that the resulting
>     multihoming would actually be useful to anyone who deployed it -
>     this is because a host-based solution only works when both
>     hosts are upgraded.  This is especially so if the scheme involves
>     rewriting applications, as well as host stacks.
> 
>     As Brian Carpenter wrote recently:
> 
>        It's clear that once you ask for action by application
>        programmers or non-routine action by end users, the costs
>        become unthinkable.
> 
> 
> 2 - As I wrote here:
> 
>      Fundamental objections to a host-based scalable routing solution
>      http://www.irtf.org/pipermail/rrg/2008-November/000233.html
> 
>     It would be undesirable to push this functionality out to hosts,
>     compared to handling it with some new architectural structures
>     in the network (that is, the routing and addressing system in
>     the core of the Net and in ISP and end-user networks).
 
I prefer a step-by-step approach. 

The first step would upgrade edge networks for multi-homing, this enables
multi-homing for hosts that can make use of it. 

In a second step, hosts are updated. It depends on the gain for users how
fast a transition takes place. Keep in mind that many end-user hosts use
dynamic addresses already (single address, IPv4, lots of NAT). Day-to-day
renumbering is no problem at all.

Upgrading server farms requires special attention. I would not say servers
MUST have static unique addresses, there are examples of hosted servers
based on DNS names. For those, multi-homing is not that difficult to
implement.


I think the subject line has some negative judgment. And I would call it the
edge-based solution.
Keep in mind that I do not prefer edge-based over core-based. I think they
are orthogonal.


Teco.



_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg

Reply via email to