[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


To: Peter Gradwell <peter@gradwell.com>
cc: dnsop@cafax.se
From: Jaap Akkerhuis <jaap@sidn.nl>
Date: Tue, 21 Nov 2000 16:20:33 +0100
In-reply-to: Your message of Tue, 21 Nov 2000 08:50:08 +0000. <5.0.0.25.0.20001121083805.0352ef18@pop3.gradwell.net>
Sender: owner-dnsop@cafax.se
Subject: Re: placement of secondary name servers for .uk.

Peter,
    
    - Currently this is done on a "traditional" and "cooperative"
    basis. The question is whether it needs to be formalised (the
    lawyers probably think so) and if so who should be asked to
    tender.

Well, we are also wondering whether we need to change this. The
current opinion seems to be that, since there is such a lot at
stake (read commercial value) on a proper working DNS, one needs
to control the ``slave'' servers as well or at least to have an
SLA with the slave server operator.

There is in CENTR an ad-hoc working group trying to come up with
what they call a shared nameserver. In this model, one registry
will run a system which will also host nameservers for others. The
operation of the hosted nameserevers can be done by the registries
themselves. The hosting registry takes only care of its own zonefile.
There is a mailing list out on this subject:

  CENTR SSS-WG mailing list <sss-wg@lists.centr.org>
  This is a closed list for the CENTR Shared Secondary Servers Working Group.
  For more information please contact <postmaster@centr.org>.

although there hardly seems to be any traffic on it.

    Currently, I'm seeking advice on the following points, and
    should be grateful for your views:
    
    a)      What is the traffic profile for DNS queries? For example,
	     should we put servers in asia? What weighting in terms
	     of numbers of servers should we give to the UK/Europe/East
	     Coast/West Coast/Mid US?

We (.nl) have currently only Europe & US covered. I assume you
could measure the traffic pattern by switching on query logging
(although this will effct perfdormance) or by sniffing. The might
give a clue. But then, the slave server operator might not like
it.
    
    - I notice that CAIDA made a report to the RSSAC, in October 1999
    http://www.caida.org/tools/measurement/skitter/RSSAC/
    
    - Does anyone know if they made any subsequent reports?

At the ICANN was again a report about the root servers. The link
is burried in the minutes (scribe notes).

    
    b)      Should DNS servers be placed on the transit LANS of
	     Exchange Points, or should they be hosted at the
	     premises of ISPs? Can they be "one hop" from an EP,
	     but at an ISP's datacentre (for ease of management).
	     What if that ISP doesn't have an open peering policy?
	     Or if they block some traffic (RBL) ?

I personal prefer an exchange point. Apart from the things you
refer to, you also avoid the sily discussion that by having it in
an ISP you would givr advanta
ge to that ISP.
    
    I understand that the root server advisory committee is asking
    similar questions for ICANN. I assume that other registries
    may be doing the same. If anyone else is currently considering
    these issues, please let me know  - perhaps we can pool brains
    and produce a combined strategy?

As said before, the centr ad-hoc workgroup is trying such a combined
strategy. And we would be happy to pool brains as well.

	jaap

Home | Date list | Subject list