[prev in list] [next in list] [prev in thread] [next in thread] 

List:       namedroppers
Subject:    Re: Implementation status and thoughts on server distribution
From:       Charles Hedrick <HEDRICK () RUTGERS ! ARPA>
Date:       1984-05-16 14:36:25
[Download RAW message or body]

Yikes!  You are certainly right.  But let me try to point out some
implications of your message that just struck me.  They are so obvious
that I am sure others must have thought about them.  But this should
still be said:

It seems fairly clear that every site with a local area network will run
a local name server.  It is clearly impractical (and probably a
violation of DCA guidelines) for our local hosts to send packets to NIC
when they want to address another local host on our local area network.
Given that situation, the protocol designers might want to think very
carefully about the requirements necessary to run a domain server. While
I have supported the idea that there should be a minimum size to have a
domain server, I now see that such a requirement poses obvious dangers.
Suppose I am in the position of being a small organization, so that I
don't have an Arpanet-sanctioned domain server.  Presumably NIC acts as
my server, or I commission someone else to do it for me.  However I still
have my own domain server, which I use on my local area network. Unless
I do something unusual, this server will also respond to packets coming
in from the Arpanet.  Here are the dangers:

  - there are now two different authoritative lists of my hosts: the one
	in the sanctioned Arpanet domain server, and the one in my local
	domain server.  Anybody who thinks those two lists will always
	be the same is dreaming.

  - anyone who communicates with my site a significant amount will be
	very tempted to put a special entry in his database to cause
	him to access my "real" domain server, rather than the 
	sanctioned one.  This will happen the first time one of his
	users wants to talk to one of my hosts that is not yet listed
	by the sanctioned domain server.

  - even if this doesn't happen, local mail will be generated using the
	local name server.

So it seems inevitable that there will be "skew" between the name
servers, and that headers in messages on the Arpanet will be generated
using both servers.  If we are careful, we will attempt to update the
sanctioned name server and the local name server at the same time, and
this skew will be minimal.  However the point of establishing size
bounds on sanctioned name servers was because we were afraid that some
sites would not be able to provide sufficient reliability.  It seems
that managing the coordination of two name servers under different
management is in fact a more serious problem than just trying to keep my
own reliable.
-------

[prev in list] [next in list] [prev in thread] [next in thread]