Re: Information & Power /Alexandria library

KPJ (kpj@sics.se)
Mon, 10 May 1999 08:41:12 +0200

It appears as if Michael S. Lorrey wrote:
|The entire DNS database is updated regularly on every web server at
|every ISP, so there is not threat to its being destroyed. Thats why they
|are called name servers. When you request a page from a site, your ISP's
|webserver looks up the DNS and its corresponding IP address, then
|queries the network for the data from that IP address. The routers take
|that query and send it to the next closer router, etc. The system does
|not query network solutions' registry every time.

The DNS (Domain Name System) data is distributed over the whole Internet, and there exists no "entire DNS database", except in the mind of the abstractor.

You could, in theory, ask every DNS server for a copy of their data base, and make a single data base of the data. In practices, however, the can be considered _impossible_:
(a) the data would be stale and incomplete at the same time:

the data change all the time everywhere. (b) the data base would require very large amounts of memory to store. (c) the bandwidth to copy it to your node would be enormous.

It appears as if Alejandro Dubrovsky <s335984@student.uq.edu.au> commented:
|
|Each DNS at each ISP holds a minuscule share of the entire database, most
|definitely not the entire database. It can cache requests that go through
|it but that only for a short time (usually one hour) and only of the
|request done on it. A DNS usually only holds the IP table for whatever
|domains it is acting as a master or as a secondary for, and usually you
|only have two servers knowing the IPs for a subdomain. Whatever ISP you
|are using does not have a clue what IP corresponds to
|s335984.student.uq.edu.au but the root servers know who the dns master for
|edu.au is (munnari.oz.au), who still doesn't know the ip but it knows that
|the master dns for uq.edu.au is krefti.uq.edu.au, and so on. This system
|enables each dns to respond quickly to name lookups since they only have
|to hold a couple of name to ip matchings but it does mean that it is
|fragile. If the 7 or 8 root servers were taken down all cross domain
|lookups (eg from .edu to .com, or from .gov to .uk) would fail after the
|corresponding caches expire (1 to 3 hours).

On May 1, 1999, there existed 13 root servers:

Name IP number Location


A.ROOT-SERVERS.NET     198.41.0.4	InterNIC (NSI) - USA
B.ROOT-SERVERS.NET     128.9.0.107	Information Sciences Institute,
					University of Southern California - USA
C.ROOT-SERVERS.NET     192.33.4.12	PSINet, Inc. - USA
D.ROOT-SERVERS.NET     128.8.10.90	University of Maryland - USA
E.ROOT-SERVERS.NET     192.203.230.10	NASA - USA
F.ROOT-SERVERS.NET     192.5.5.241	Internet Software Consortium - USA?
G.ROOT-SERVERS.NET     192.112.36.4	Network Information Center - USA
H.ROOT-SERVERS.NET     128.63.2.53	U.S. Army Research Laboratory,
					Maryland - USA
I.ROOT-SERVERS.NET     192.36.148.17	NORDUnet - Scandinavia
J.ROOT-SERVERS.NET     198.41.0.10	InterNIC (NSI) - USA
K.ROOT-SERVERS.NET     193.0.14.129	RIPE Network Coordination Centre (NCC)
					Amsterdam - Netherlands
L.ROOT-SERVERS.NET     198.32.64.12	ISI (IANA) - USA
M.ROOT-SERVERS.NET     202.12.27.33	Widely Integrated Distributed
					Environment (WIDE) - Japan

A list of the current DNS root servers can be acquired from <URL:ftp://RS.INTERNIC.NET/domain/named.root>, which might be a good idea for all DNS server administrators to do now and then.