On Tue, 2002-11-26 at 18:06, Keith Moore wrote: > One difference between our models may be that you seem to be assuming > that if a network has external connectivity, it has connectivity to > the public Internet.
Your right. I have been assuming that external = public Internet. But I have also assumed that internal != public Internet. If two organisations decide to connect together via a backdoor connection, push GUPI routes between each other's routing domain, and implement two opposite facing firewalls, I would still consider their connection internal. I do not assume that. So I see GUPIs as a way > in which networks that aren't connected to the public Internet can > still get addresses which allow them to establish private connections > to external networks. With my definitions of internal / external above, I'm struggling to think of an external network that wouldn't be the Internet. I can imagine a service provider / telco building a private IPv6 network for their customer's to connect to, which I think could be an "external" network using your definition. OTOH under my definition of internal above, this private telco backbone could also be considered internal - you would be connecting to a network which provides you with connectivity to other, known parties that are part of this private IPv6 backbone. It is a conscious decision to obtain connectivity with these other parties. The service provider is only really providing convenient, managed layer 3 connectivity, and ensuring better QOS, by providing dedicated bandwidth to known customers. I suppose basically I'm considering internal to be any time one organisation chooses to make its GUPI address space routes available to another, and accept the other organisation's GUPI address space routes. The organisation knows who it is talking to and vice versa (I'm not talking about a trust relationship here though - just the conscious decision to interconnect to each other's networks and the trading of GUPI routes). External under my definition is when you choose to connect to a network, where you don't know and can't control who else is connected to it - the public Internet is the only network which I know of that fits this definition. > > Another difference is that I see little reason for a network to support > both GUPIs and globals - if a network has globals then it is probably > better off without GUPIs. Yes, GUPIs might be more stable than globals, > but this is not necessarily the case - the opposite could also be true. > Other than internal links failing, causing destination unreachables, timeouts etc, what cases would there be where globals are more stable that GUPIs ? I'm struggling to think of a failure case for the new GUPI models where stability would be less than globals. Renumbering when merging networks was a stability issue we had with traditional site-locals, but that is the what GUPIs fixes. I imagine that from the day you switch on your first router and it generates / you assign your GUPI addressing, your internal GUPI addressing would stay the same as you build out your network. Internal communications failures would be caused by all the other typical causes of failure in the network, but not GUPI addressing itself. > OTOH I wouldn't try to insist that a network that supports globals > not use GUPIs. If supporting both kinds of prefixes turns out to be > the best way for that particular network to operate, fine. I just > don't see that configuration as being of signifcant value. > I think there is, but at least they are optional - I would use them, and you wouldn't :-) Regards, Mark. -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] --------------------------------------------------------------------
