Re: A new spin on multihoming: multihoming classes.

[Peter Tattam <peter@jazz-1.trumpet.com.au>]

On Fri, 7 Sep 2001, Geoff Huston wrote:

> At 9/7/01 01:02 AM, RJ Atkinson wrote:
> 
> But as to the assertion that 8,192 is some magical preferred number of 
> 'root prefixes', then I'm not sure that I can agree.
> 

Geoff,

I am concerned that the whole multihoming issue hinges on the answer to whether
BGP can be made to work with larger DFZ than we anticipated.

I'm not sure where 8K comes from.  8K = 2^13. 

Maybe it relates to the size of a practical switching table in very fast
routers built using ASICs. 

Maybe it's just a figure which was plucked out of the air in relation to the
BGP table size at the time of the IPv6 standards being formed.

Maybe it's a figure designed to stabilize the core given the high bandwidths
we're likely to experience in the future.  We have probably gotten to the point
where we finally have enough capacity on inter site links and now RTT delays
will be governed

My impression however is that it has been generally agreed (amongst Ipv6 people
at least) that BGP in its current form isn't going to scale.

If I asked you the question "how big is too big" what would you answer, and how
would base your measure?

My understanding wasn't that it was hardware size limitations that we critical,
but more that the amount of information grows faster than O(n) resulting in
exchange of that information causing instability of the DFZ.  

In my opinion some of the measures that prevent route flapping may actually
cause more damage than good because they introduce stepping functions into the
differential equation that would model the DFZ.  They work now because the
paramters have been determined empirically - I'm uncertain that they will scale
indefinitely.

Peter

--
Peter R. Tattam                            peter@trumpet.com
Managing Director,    Trumpet Software International Pty Ltd
Hobart, Australia,  Ph. +61-3-6245-0220,  Fax +61-3-62450210