>it's not at all clear that this can work well enough to be a general
>purpose multihoming mechanism, at least not without adding a fair
>amount of extra infrastructure and complexity - i.e. a mechanism
>which hosts or applications can use to query the network to determine
>relative proximity of several different addresses.
There is also a potential scaling issue of using multiple addresses
as general purpose multihomging mechanism. This is because if this
is the case, most of the Internet hosts will end up with multiple
addresses.
Based on the current assignment strategy, only top-tier ISPs (networks)
will be assigned with TLAs. tier-2 ISPs will get allocation address
space from the TLAs of their connected tier-1 ISPs. If a tier-2 ISP
is multihomed, the ISP will be assigned with two IP addresses from two
different TLAs. All the downstream customers of this tier-2 ISP will
get multiple addresses for each host. Since most of the tier-2 ISP
(if not all) will be multihomed to tier-1 ISPs for redundancy purpose,
most of the Internet hosts will end up with multiple IP addresses.
Only tier-1 ISPs' direct connected single-homed customers will not
have multiple addresses for their hosts.
It's possible that some multihomed sites will have to assign 4 or even
more ip addresses per host, depend on what kind of ISPs they multihoming
with. E.g. a site that happen to multihome to two tier-2 ISPs, each
multihomed with two different tier-1 ISPs, each host in this multihomed
site will have 4 IP addresses in order to get full benefit of redundancy.
One can image even more complicated case.
So I think we need to explore different ways of doing multihoming.
--Jessica
Date: Thu, 09 Dec 1999 19:36:28 EST
To: Christian Huitema <[EMAIL PROTECTED]>
cc: Sean Doran <[EMAIL PROTECTED]>, [EMAIL PROTECTED]
From: Keith Moore <[EMAIL PROTECTED]>
Subject: Re: IP network address assignments/allocations information?
Return-Path: [EMAIL PROTECTED]
X-URI: http://www.cs.utk.edu/~moore/
In-reply-to: Your message of "Thu, 09 Dec 1999 18:14:18 EST."
<[EMAIL PROTECTED]>
Sender: [EMAIL PROTECTED]
Content-Type: text
Content-Length: 1465
> Let's put it this way: the registries are instructed that only top level
> providers should get one of these addresses. Everyone who does not qualify
> supposedly get a delegation from a TLA, or several delegations in the case
> of multi-homed networks.
of course, this requires that sending hosts or applications make
intelligent decisions about which destination address to use
(and which source address to use with a particular destination
address), usually in the absence of any information which might
inform the decision.
it's not at all clear that this can work well enough to be a general
purpose multihoming mechanism, at least not without adding a fair
amount of extra infrastructure and complexity - i.e. a mechanism
which hosts or applications can use to query the network to determine
relative proximity of several different addresses. if it does turn
out to work it will probably be because all of the available prefixes
for both the source and destination host are so reliable and have so
much available bandwidth that most of the time that it doesn't matter
which of the available addresses you use. (it's tempting to say that
multihoming will work quite well for those cases where you don't need
multihoming... but that is a bit of an exaggeration)
to be fair, "traditional" multihoming doesn't scale well enough
to use that approach either.
maybe we just need for Internet service to be as reliable as telephone
service.
Keith
