Non-ISP companies multi-homing?

Fri Jul 25 20:31:16 UTC 1997

On Fri, 25 Jul 1997, Eric Germann wrote:

> I completely subscribe to your option theory, however, I've been told and
> seen ping sweeps to see if the space is in use.  And the other response is

If you ping my my class B, or indeed my /23 you'll get back a host 
unreachable with a type 13 (administrativel prohibited), even for host 
addresses which are legitimately routable, so that's not really a valid 
test.  If you're overly agressive, you'll probably also get a phone call.

> if you don't want them seen now, give them 10.x or 192.168, and use a
> proxy.  So it doesn't fall under the NIC's and presumably ARIN's
> allocations policies.

When we registered the addresses, we didn't have a firewall.  We were 
pretty much without clue, and the "plan" as it were, from the group which 
handled it at that time was to be able to selectively address machines as 
the need arised.

If tomorrow, I decided to start hosting Web services for all of my 
business units, I'd give a *lot* of established server farms a heck of a 
run, and I'd need more than a /23 to do it.  

There are also things that proxies don't scale to. 

> probably a /23 at most.  Using reponsible and aggressive management of
> allocation policies, they will grow up within a year or two to a /19 or
> larger.  But their early customers are SOL for anything on the far side of
> Sprint, unless of course, we pay Sprint.  And every other NSP who has a /19
> filter in place.  So we can't multihome, buy transit from a couple of good
> NSP's and let the economics drive our decision.

I'm really curious if anyone has thought this through with all the VPN, 
"plug in anywhere", and addressable atoms we're promised in IPv6?  

This also depends on the NSPs, BBN and UUNet both didn't have a problem 
routing my /23s, and I'd initially come to the table thinking that I'd be
stuck with only my class B, of which only one subnet sits outside.  I'm 
not sure how they'd have handled it if the /23 had come from their 
address space though.

> So getting in early is a good thing I suppose.  Day late and a dollar short
> for the rest of 'em.

Yeah, unfortunately IPv6 (if and when) will mean a stampede.  Does anyone 
have any experience with routing tables under v6 yet?   If you want to, 
drop me a private note, as this is probably getting out of NANOG.

Paul
-------------------------------------------------------------------------
Paul D. Robertson
gatekeeper at gannett.com