To:
Masataka Ohta <mohta@necom830.hpcl.titech.ac.jp>
cc:
hobbes@engin.umich.edu (Steve Mattson), dnsop@cafax.se
From:
marka@isc.org
Date:
Thu, 08 Jul 1999 09:52:15 +1000
In-reply-to:
Your message of "Thu, 08 Jul 1999 07:53:16 +0200." <199907072253.HAA13316@necom830.hpcl.titech.ac.jp>
Sender:
owner-dnsop@cafax.se
Subject:
Re: Topological Motivation for draft-ohta-root-servers-01.txt?
> Steve Mattson; > > > Upon reading it, and previous versions of same, I have noticed the > > repeated country-specific references as motivation for the draft. > > However, it appears that the operational changes proposed in the > > draft have very little to do with national borders but instead with > > network topology issues involving adjacent AS's and BGP routing. > > It reflects the fact that most of the current root servers are in US > and people in US, seemingly, do not so much feel a lot of root > servers necessary. > > > I believe that a reworded motivation section of this draft based solely > > on network topology concerns would allow for a debate centered on the > > issue at hand. > > The description explain the motivation. Abstract wording backed up by > no real-world situation it useless. > > It is a lot better to remove it totally than make it abstract. > I suspect that it would be better to say that the servers need to be distributed so that there is, in general, not the need to make inter-continental, trans-oceanic or satellite hops. All of these types of links introduce large delays. I know there a political reasons at times to make the servers closer, but I believe the above critera technical critera should be used for determining how many root servers there are sharing a given unicast address and there rough placement. The driving factor should be response times given an otherwise unloaded network. Does < 100ms sound like a good figure? While I don't know what the Atlantic adds to the RTT. I do know that satellites add ~400ms and the Pacific ~200ms. From an experimental point, BIND uses bands 128ms wide with the smallest band 64ms wide to select which nameserver to choose and it does a reasonable job of keeping named talking to the closest servers. The size of these bands is adjustable at compile time. > > This would allow the draft to proceed without having to > > address what is "internationally fair" or how to determine an evolving > > weight factor for the default set of root servers which one overrides > > selectively within one's "local" Internet. > > What do you mean "local" Internet? > > The proposal is for the global Internet. > > > At the very least a statement > > could be added to this i-d which specifically declared the determination > > of the default set of root servers, or the default scope of those root > > server addresses not otherwise overridden, as being beyond the scope of > > the draft. > > It is determined to be an empty set. > > The i-d specifically declares that all the root server addresses can > be overridden. > > > Apologies if I've only added to the current confusion, > > Don't mind. I certainly hate confusion but love chaos. > > Masataka Ohta > -- Mark Andrews, Internet Software Consortium 1 Seymour St., Dundas Valley, NSW 2117, Australia PHONE: +61 2 9871 4742 INTERNET: marka@isc.org