Re: [narten@us.ibm.com: PI addressing in IPv6 advances in ARIN]

["Per Heldal" <heldal@eml.cc>]

On Fri, 14 Apr 2006 13:12:42 +0200, "Iljitsch van Beijnum"
<iljitsch@muada.com> said:
> On 14-apr-2006, at 12:42, Per Heldal wrote:
> 
> >> An alternative would be to give the prefixes out based on geography,
> >> so there is at least a chance that we get to aggregate them
> >> geographically in the future. But despite the fact that doing this is
> >> entirely risk-free (geo PI without geo aggregation is no worse than
> >> regular PI) people want proof that it works first, while at the same
> >> for deploying regular PI the burden of proof is reversed so it's
> >> allowed now despite very good indications (but not hard proof) that
> >> it will be problematic in the future.
> 
> > Geo-aggregation doesn't work in a competitive market. It would require
> > new routing-technology to do so, and thus fall in the same category as
> > shim6.
> 
> Well obviously you don't know what you're talking about (not  
> apologizing for my crankiness, btw) or we wouldn't be having this  
> discussion in the first place.

The alternative to new technology is to break the business-model for
transit. Endless treads on this issue over 10 or more years have mainly
concluded that with existing technology transit providers either have to
provide equal connectivity to everyone in a region (customer as
non-customer), or eliminate the gain from geotop by announcing
more-specific routes. This problem currently has no technical solution
... hence something else is needed.

[snip]
> 
> The challenge for the community as a whole is to accept limitations  
> rather than ignore them and hope the problems that are created by  
> behaving that way will magically disappear.

It's not the first time in history researchers fail to meet deadlines or
otherwise have to accept that society moves on given whatever resources
are available at the time. 

> 
> The IETF been talking about scalable routing = scalable multihoming  
> for at least a decade and it's entirely clear that there is no pain- 
> free solution that allows all the current benefits without incurring  
> any of the bad side effects. Ignore this lesson at your peril.

In some distant history (pre 1990) the IETF would just have moved on
making a practical solution based on actual requirements. Now it's about
to become a second ITU,  hampered by procedures and unable to make a
decision in any issue where there is more than one opinion. 


> 
> >> IPv6 is really just repeating the same mistake with longer addresses.
> 
> > I didn't say it's a perfect solution, but we're not living in a  
> > perfect
> > world.
> 
> Oh, my mistake. Obviously because the world is imperfect it's ok to  
> do things we know are dumb beforehand.
> 
> > If we're unable to come up with a product the market wants we
> > have to accept that the market makes the most of what it's got
> > regardless of what we may think about it.
> 
> Yeah, ain't capitalism grand. To quote 10cc:
> 
> To do the Wall Street Shuffle
> Let your money hustle
> Bet you'd sell your mother
> You can buy another

This isn't necessarily about money. To some it's about reality vs
fantasy. 

IPv6 without PI, shim6 or some other alternative won't fly.

However, an alternative is to stop V6 deployment until the supporting
technology is ready ... if you can hold back the floodgates of hype
driving it.

> 
> > PS! I believe the probability of something entirely new replacing
> > IPv<anything> is much greater than the probability of the internet
> > falling apart due to its growth.
> 
> Say hi to the easter bunny if you run into it this weekend, by the  
> way. Belief has no place in engineering. But it doesn't matter one  
> way or the other, as I'm sure the ARIN constituents will reject it if  
> it's really this great.

Turning to reality; I haven't seen anything indicating that practical
limitations have come closer in recent years. In fact, the average
router today has more headroom than its ancestors back in '98 or '99.

Demonstrate how to implement scalable shim6 with traffic engineering in
a network with 10k hosts of various types this month, and I'll stop
talking about PI.

//per
-- 
  Per Heldal
  http://heldal.eml.cc/