To:
George Belotsky <george@register.com>
cc:
Bill Manning <bmanning@ISI.EDU>, Shane Kerr <shane@ripe.net>, ietf-provreg@cafax.se, ietf-whois@imc.org
From:
Sheer El-Showk <sheer@laudanum.saraf.com>
Date:
Mon, 29 Jan 2001 11:01:32 -0500 (EST)
In-Reply-To:
<20010129103932.A735@register.com>
Sender:
owner-ietf-provreg@cafax.se
Subject:
Re: Merging RRP and Whois
> > % 1. The ProvReg group designs a protocol. This protocol allows/assumes: > > % * A centralized object repository (registry) is assumed. > > > > Why is this assumption in place? > > One could (rightly) argue that the single largest cause of > > instability and scaleability is the insistance on using > > "A centralized ... repository". The problems with that Well, I wouldn't say this is necassarily true. DNS is extremely distributed, both in location and in authority (well ICANN on top, but at lower levels everyone only sees the servers above them as local authority, they don't all go straight to the root servers) and that has not helped make it any more stable (it just means you can have that many more sources of inaccuracy -- the authoritative domain server, a glue record in a higher level domain, a none-updated record in a secondary server). As for scalability of a centralized repository ... it shouldn't be a problem. Centralized doens't mean one server or one network ... it means one central authority. They can use as many redundant, load-balancing servers as they want -- the idea is that there is always one entity to turn to to get authoritative information regarding a domain or other registered object. > > tactic caused the original IR to segment into multiple > > regional IRs, each retaining/maintaining "A centralized > > repository". Its gotten worse with the addition of each new > > "routing database" & whois service by agency. Each presumes > > a single "centralized repository". > > > > I'd rather see a protocol to allow a composite, non authoritative > > structure be fabricated from collections of hundreds/thousands > > of broadly distributed attributes. That way I would own my > > data and be able to direct its distribution to/through others > > non-auth copies of my data. This sounds nice in principle, but how do you maintain data consistency, data-uniqueness, and how do you protect against data corruption. I'd be genuinly interested in discussing a system that would allow the data to be maintain its own integrity in some way and be distributed (eg just an interesting idea: unique world-wide NIC handle with your region specific NIC info being held in the ccTLD registries keyed to your unique identifier with local privilage systems in place -- sounds nice but a little complicated). Sheer