Sorry, I thought we were discussing all multi-homing. Your example doesn't
help the business whose ISP suffers a business failure (such as DSLnetworks,
Flashpoint, et al), only the case where the access provider fails SLA.
To put it bluntly, a single circuit, to a single business, while annoying,
doesn't cause wide-spread outages when it fails. It is thus, the
lesser-order case. The case I thought was under discussion is when an ISP
dumps something on the order of 10^3 or more customers when they fail. I
understand that NorthPoint abandoned
~100,000 customers when they sold their backbone to AT&T and AT&T didn't
pick up the subscribers. I will wager that many of them were /24s.
DSLnetworks had over 700 Covad customers, FlashPoint was larger. For various
definitions of "wide-spread", this is a much larger issue than a broken
copper-pair. I suspect that it also has a much higher likelyhood of
occurance. Especially, in the current business shakeout.
Guess what ... it won't stop. This sort of problem will be with us forever.
We should find a solution ... someday ... ya think?
Business failures are on one side of the problem and CIDR aggregation is on
the other.
Northpoint was an access provider, not really an ISP.
Folks using Northpoint had someone else as their ISP. I tried
last year to use Northpoint as an ISP and they said, sorry,
we aren't an ISP, here is our list of ISP partners, pick one.
If customers had a 2nd access circuit from a different access
provider to a different POP with the same ISP, they would be fine
(as my friend has been in just this situation) even when
Northpoint turned off the DSL access line.
What AT&T bought wasn't so much a backbone as a whole
lot of DSLAMs that are pre-located inside RBOC facilities
and pre-interconnected with the RBOC facilities. AT&T
already had at least 2 backbones (one obtained from TCG/CERFnet
and one from WorldNet), hence didn't need another.
Quod Est Demonstratum.
Ran
what if an isp with a lot of dedicated line customers in old A space takes
the dive and a few thousand of their customers come to isps each asking to
have 38.42.666.0/24 or whatever routed?
mirjam says
o route them now, forcing instant renumbering would add insult to injury
o give them the normal 90 days to renumber
and she's never led me wrong so far.
randy
what if an isp with a lot of dedicated line customers in old A space
takes
the dive and a few thousand of their customers come to isps each asking
to
have 38.42.666.0/24 or whatever routed?
mirjam says
o route them now, forcing instant renumbering would add insult to
injury
o give them the normal 90 days to renumber
and she's never led me wrong so far.
randy
In the example cited, the only obstacle would be registry authority. (After
all, in theory as soon as the registry gets official notification and the
block is no longer advertised, it could be reassigned.) So the fact that
Mirjam said so automatically makes it so.
But consider a similar case where an ISP 'sort of' goes out of business.
Perhaps they're still BGP advertising their main block, but they've stopped
providing service to a customer and there's a dispute as to whether the
customer was 'cut off' or not. As soon as the ISP cuts off the authority,
doesn't the customer no longer have any claim to (route) the block?
Isn't this (routing table growth and arguments over who 'really owns' an IP
address block) the reason subassigned IP addresses are (now) non-portable?
DS