From kerr at arin.net  Wed Apr 26 16:19:05 2000
From: kerr at arin.net (Shane Kerr)
Date: Wed, 26 Apr 2000 16:19:05 -0400 (EDT)
Subject: Maximum number of DNS servers
Message-ID: <Pine.LNX.4.10.10004261615160.19727-100000@helium.arin.net>

I'm working on some of the requirements for the new system, and I had a
quick question:  we can either use a 1 or 2 digit number to order
IN-ADDR.ARPA servers in our database.  Can anyone think of a reason to
need more than 10 servers?

--
Shane Kerr <kerr at arin.net>
Senior Software Engineer
American Registry for Internet Numbers (ARIN)
+1 703-227-9877


From Tanya.Hinman at cwusa.com  Wed Apr 26 17:25:48 2000
From: Tanya.Hinman at cwusa.com (Tanya Hinman)
Date: Wed, 26 Apr 2000 17:25:48 -0400
Subject: FW: Maximum number of DNS servers
Message-ID: <001201bfafc5$fd2769e0$242547cc@thinman.cary.cw.net>


-----Original Message-----
From: Manny Kontos [mailto:emmanuel.kontos at cwusa.com] 
Sent: Wednesday, April 26, 2000 5:09 PM
To: Tanya.Hinman at cwusa.com
Subject: RE: Maximum number of DNS servers


No, the minimum requirement is two, for redundancy and back-up.  Anything
over this is for traffic distribution and maximum
fail-over.  Anything over 4 or 5 is unnecessary and overkill........

Manny

 -----Original Message-----
From: 	Tanya Hinman [mailto:Tanya.Hinman at cwusa.com]
Sent:	Wednesday, April 26, 2000 4:58 PM
To:	Manny Kontos
Subject:	FW: Maximum number of DNS servers

Manny,

I thought you might be interested in the below email from ARIN.

Tanya

-----Original Message-----
From: dbwg-request at arin.net [mailto:dbwg-request at arin.net] On Behalf Of
Shane Kerr
Sent: Wednesday, April 26, 2000 4:19 PM
To: dbwg at arin.net
Subject: Maximum number of DNS servers


I'm working on some of the requirements for the new system, and I had a
quick question:  we can either use a 1 or 2 digit number to order
IN-ADDR.ARPA servers in our database.  Can anyone think of a reason to
need more than 10 servers?

--
Shane Kerr <kerr at arin.net>
Senior Software Engineer
American Registry for Internet Numbers (ARIN)
+1 703-227-9877


From jerry at fc.net  Wed Apr 26 17:59:19 2000
From: jerry at fc.net (Jeremy Porter)
Date: Wed, 26 Apr 2000 16:59:19 -0500
Subject: Maximum number of DNS servers 
In-Reply-To: Your message of "Wed, 26 Apr 2000 17:35:11 EDT."
             <20000426173511.A22485@ultra.net> 
Message-ID: <200004262159.QAA94066@freeside.fc.net>



Unfortunately that is how SQL database are designed, as per the spec.
While logically an 8 bit value would be fine, the database specification
language defines it in decimal digits, or 3.5 bits.
Since this constraint exists one needs to consider it.

I can't see any realistic need for more than 9 name servers
for in-addr.arpa, since  in-addr.arpa usage is a secondary function
and not an area where people are even trying to do funky things with
dns.


In message <20000426173511.A22485 at ultra.net>, Joe Provo - Network Architect writes
:
>On Wed, Apr 26, 2000 at 04:19:05PM -0400, Shane Kerr wrote:
>> I'm working on some of the requirements for the new system, and I had a
>> quick question:  we can either use a 1 or 2 digit number to order
>> IN-ADDR.ARPA servers in our database.  Can anyone think of a reason to
>> need more than 10 servers?
>
>I'm not certain if any studies have been made regarding any "ceilings of
>effectiveness" for NS delegation.  Barring any such technical difficulty
>or BCP quoting a number beyond which it becomes stupid to add more, I'm
>stunned that such an arbitrary limit would even be considered.  Should I
>ask why the database system case how many digits [character
>representations] there are rather than a true value [bit-size of the
>actual value].  
>
>Just because I can't think of a need for more than 5-7 for a given
>netblock doesn't mean someone else won't.  If you have to make this kind
>of arbitrary, display-oriented (rather than value-oriented) cutoff,
>Internet experience dictates to go with the larger of your choices.  
>
>Cheers,
>
>Joe
>--
>Joe Provo                                            Voice  508.486.7471
>Director, Internet Planning & Design                 Fax    508.229.2375
>Network Deployment & Management, RCN                 <joe.provo at rcn.com>
>

--- jerry at fc.net
Director Network Operations/Network Engineering, Wayport, Inc.
512-519-6193 www.wayport.net
8303 Mopac Expressway Suite A300, Austin Tx.


From repete at cncx.com  Wed Apr 26 18:13:54 2000
From: repete at cncx.com (Pete Bowden)
Date: Wed, 26 Apr 2000 15:13:54 -0700 (PDT)
Subject: Maximum number of DNS servers
In-Reply-To: <Pine.LNX.4.10.10004261615160.19727-100000@helium.arin.net> from "Shane Kerr" at Apr 26, 2000 04:19:05 PM
Message-ID: <200004262213.PAA21217@skylab.eng.internex.net>

I thought there was a limit of 9 specified in the RFC's, but I'm probably
confused...  Perhaps the solution is to allow for that to be alpha-numeric
and if you need more than 10 you can start with the letters A-Z :-)


> 
> I'm working on some of the requirements for the new system, and I had a
> quick question:  we can either use a 1 or 2 digit number to order
> IN-ADDR.ARPA servers in our database.  Can anyone think of a reason to
> need more than 10 servers?
> 
> --
> Shane Kerr <kerr at arin.net>
> Senior Software Engineer
> American Registry for Internet Numbers (ARIN)
> +1 703-227-9877
> 


-- 
Pete Bowden, Internet Network Engineer, Internet & Data Center Engineering
    rePete at concentric.com  rePete at cncx.com  pete at internex.net  NIC:PB8
Concentric Network Corporation, 1400 Parkmoor Ave., San Jose, CA  95126-3429
		  Voice: 408-808-6010    Fax: 408-808-6010


From joe.provo at rcn.com  Wed Apr 26 17:35:11 2000
From: joe.provo at rcn.com (Joe Provo - Network Architect)
Date: Wed, 26 Apr 2000 17:35:11 -0400
Subject: Maximum number of DNS servers
In-Reply-To: <Pine.LNX.4.10.10004261615160.19727-100000@helium.arin.net>; from kerr@arin.net on Wed, Apr 26, 2000 at 04:19:05PM -0400
References: <Pine.LNX.4.10.10004261615160.19727-100000@helium.arin.net>
Message-ID: <20000426173511.A22485@ultra.net>

On Wed, Apr 26, 2000 at 04:19:05PM -0400, Shane Kerr wrote:
> I'm working on some of the requirements for the new system, and I had a
> quick question:  we can either use a 1 or 2 digit number to order
> IN-ADDR.ARPA servers in our database.  Can anyone think of a reason to
> need more than 10 servers?

I'm not certain if any studies have been made regarding any "ceilings of
effectiveness" for NS delegation.  Barring any such technical difficulty
or BCP quoting a number beyond which it becomes stupid to add more, I'm
stunned that such an arbitrary limit would even be considered.  Should I
ask why the database system case how many digits [character
representations] there are rather than a true value [bit-size of the
actual value].  

Just because I can't think of a need for more than 5-7 for a given
netblock doesn't mean someone else won't.  If you have to make this kind
of arbitrary, display-oriented (rather than value-oriented) cutoff,
Internet experience dictates to go with the larger of your choices.  

Cheers,

Joe
--
Joe Provo                                            Voice  508.486.7471
Director, Internet Planning & Design                 Fax    508.229.2375
Network Deployment & Management, RCN                 <joe.provo at rcn.com>


From danny at tcb.net  Wed Apr 26 23:57:27 2000
From: danny at tcb.net (Danny McPherson)
Date: Wed, 26 Apr 2000 21:57:27 -0600
Subject: Maximum number of DNS servers 
Message-ID: <200004270357.VAA03975@tcb.net>


I agree.  If choosing the larger value doesn't impose any additional 
constraint then it would certainly be the optimal solution.  It's not
as if you're attempting to reduce protocol overhead or the like.

-danny

> I'm not certain if any studies have been made regarding any "ceilings of
> effectiveness" for NS delegation.  Barring any such technical difficulty
> or BCP quoting a number beyond which it becomes stupid to add more, I'm
> stunned that such an arbitrary limit would even be considered.  Should I
> ask why the database system case how many digits [character
> representations] there are rather than a true value [bit-size of the
> actual value].  
> 
> Just because I can't think of a need for more than 5-7 for a given
> netblock doesn't mean someone else won't.  If you have to make this kind
> of arbitrary, display-oriented (rather than value-oriented) cutoff,
> Internet experience dictates to go with the larger of your choices.  
> 
> Cheers,
> 
> Joe
> --
> Joe Provo                                            Voice  508.486.7471
> Director, Internet Planning & Design                 Fax    508.229.2375
> Network Deployment & Management, RCN                 <joe.provo at rcn.com>


From kerr at arin.net  Wed Apr 26 18:40:18 2000
From: kerr at arin.net (Shane Kerr)
Date: Wed, 26 Apr 2000 18:40:18 -0400 (EDT)
Subject: Maximum number of DNS servers 
In-Reply-To: <200004270357.VAA03975@tcb.net>
Message-ID: <Pine.LNX.4.10.10004261802560.19822-100000@helium.arin.net>

I made a quick scan of the ARIN database, and here are the counts of
number of hosts for networks:

0 hosts = 343044
1 host  = 359
2 hosts = 26630
3 hosts = 5646
4 hosts = 2042
5 hosts = 641
6 hosts = 165
7 hosts = 10
8 hosts = 1

(Not sure about those 359 entries with one host.  There's not much we can
do about those entries, except perhaps contact the adminstrators and ask
for clean up.)

Currently the list of host handles must fit in a fixed-size field of the
database, so the number of hosts is limited, but it depends on the actual
handles of the host records (that is, A-HST and B-HST might fit where
SOMEOTHER-HST might not).

Choosing the smaller limit does save us a couple of Kbyte of space, but
what I'm really thinking there are perhaps user interface designs that are
practicable with sets of 10 that aren't really feasable with sets of 100.
The reason the number 10 came up is that it's handy in an SQL database,
but the advantage lies in dealing with the group.  If we have a smallish
limit, we can build HTML forms and such around that limit.  If there is no
such limit, then a more sophisticated (and possibly cumbersome and
confusing) interface needs to be designed:

     +--------------------------------+
  1. | NS.FOO.BAR.COM                 |
     +--------------------------------+
     +--------------------------------+
  2. | NS2.FOO.BAR.NET                |
     +--------------------------------+
     +--------------------------------+
  3. | NS3.FOO.BAR.ORG_               |
     +--------------------------------+
  
       +------------+   +------+
       | Add Server |   | Done |
       +------------+   +------+

The user would either need to run a fairly complicated JavaScript program
or send an HTML form to the server multiple times to build a list.  Not
that this is necessarily so bad, but there it is.  Of course, the
interfaces can have limits that the database itself does not.

>From a programmers perspective, there are only 3 numbers: 0, 1, and many.
:)

--
Shane Kerr <kerr at arin.net>
Senior Software Engineer
American Registry for Internet Numbers (ARIN)
+1 703-227-9877

On Wed, 26 Apr 2000, Danny McPherson wrote:

> I agree.  If choosing the larger value doesn't impose any additional
> constraint then it would certainly be the optimal solution.  It's not
> as if you're attempting to reduce protocol overhead or the like.
> 
> -danny
> 
> > I'm not certain if any studies have been made regarding any "ceilings of
> > effectiveness" for NS delegation.  Barring any such technical difficulty
> > or BCP quoting a number beyond which it becomes stupid to add more, I'm
> > stunned that such an arbitrary limit would even be considered.  Should I
> > ask why the database system case how many digits [character
> > representations] there are rather than a true value [bit-size of the
> > actual value].  
> > 
> > Just because I can't think of a need for more than 5-7 for a given
> > netblock doesn't mean someone else won't.  If you have to make this kind
> > of arbitrary, display-oriented (rather than value-oriented) cutoff,
> > Internet experience dictates to go with the larger of your choices.  
> > 
> > Cheers,
> > 
> > Joe
> > --
> > Joe Provo                                            Voice  508.486.7471
> > Director, Internet Planning & Design                 Fax    508.229.2375
> > Network Deployment & Management, RCN                 <joe.provo at rcn.com>



From markk at netsol.com  Wed Apr 26 20:32:36 2000
From: markk at netsol.com (Mark Kosters)
Date: Wed, 26 Apr 2000 20:32:36 -0400
Subject: Maximum number of DNS servers
In-Reply-To: <Pine.UW2.4.10.10004270914160.19872-100000@compiler.staff.apnic.net>
References: <Pine.LNX.4.10.10004261615160.19727-100000@helium.arin.net> <Pine.UW2.4.10.10004270914160.19872-100000@compiler.staff.apnic.net>
Message-ID: <20000426203236.B6404@slam.internic.net>

Paul is right - you'll need two digits. Very few servers 10 or more
but those who do should not be constrained by some limitation in the
database. Plus when EDNS comes out, non-secure servers may want more.

Mark

On Thu, Apr 27, 2000 at 09:21:22AM +1000, Paul Gampe wrote:
> The response to a DNS query for NS records will need to fit into a udp
> packet which limits you to 512 octets. [rfc1035]  The root servers have
> the same constraint and cannot exceed 13, so it would be highly unlikely
> for you to need more then 2 digits I imagine.

-- 

Mark Kosters             markk at netsol.com       Network Solutions, Inc.
PGP Key fingerprint =  1A 2A 92 F8 8E D3 47 F9  15 65 80 87 68 13 F6 48


From dploher at level3.net  Wed Apr 26 23:07:52 2000
From: dploher at level3.net (Darren Loher)
Date: Thu, 27 Apr 2000 03:07:52 +0000
Subject: Maximum number of DNS servers
In-Reply-To: <20000426203236.B6404@slam.internic.net>; from markk@netsol.com on Wed, Apr 26, 2000 at 08:32:36PM -0400
References: <Pine.LNX.4.10.10004261615160.19727-100000@helium.arin.net> <Pine.UW2.4.10.10004270914160.19872-100000@compiler.staff.apnic.net> <20000426203236.B6404@slam.internic.net>
Message-ID: <20000427030752.F29100@zed.eng.level3.com>

Level 3 has 30 DNS servers that are authoritative for our
in-addr.arpa domains.  But they all share just two IP addresses. 
:)

I also think that using 2 digits is a good idea.  It is a small
amount of overhead in your SQL database.

-Darren

-- 
Darren Loher                 Level 3 Communications
dploher at level3.net           Global Data Architecture
720-888-2847 (office)        "Sed quis custodiet ipsos custodes?"

On Wed, Apr 26, 2000 at 08:32:36PM -0400, Mark Kosters wrote:
> Paul is right - you'll need two digits. Very few servers 10 or more
> but those who do should not be constrained by some limitation in the
> database. Plus when EDNS comes out, non-secure servers may want more.
> 
> Mark
> 
> On Thu, Apr 27, 2000 at 09:21:22AM +1000, Paul Gampe wrote:
> > The response to a DNS query for NS records will need to fit into a udp
> > packet which limits you to 512 octets. [rfc1035]  The root servers have
> > the same constraint and cannot exceed 13, so it would be highly unlikely
> > for you to need more then 2 digits I imagine.
> 
> -- 
> 
> Mark Kosters             markk at netsol.com       Network Solutions, Inc.
> PGP Key fingerprint =  1A 2A 92 F8 8E D3 47 F9  15 65 80 87 68 13 F6 48


From paulg at apnic.net  Wed Apr 26 19:21:22 2000
From: paulg at apnic.net (Paul Gampe)
Date: Thu, 27 Apr 2000 09:21:22 +1000 (EST)
Subject: Maximum number of DNS servers
In-Reply-To: <Pine.LNX.4.10.10004261615160.19727-100000@helium.arin.net>
Message-ID: <Pine.UW2.4.10.10004270914160.19872-100000@compiler.staff.apnic.net>

On Wed, 26 Apr 2000, Shane Kerr wrote:

kerr>I'm working on some of the requirements for the new system, and
kerr>I had a quick question:  we can either use a 1 or 2 digit number
kerr>to order IN-ADDR.ARPA servers in our database.  Can anyone think
kerr>of a reason to need more than 10 servers?

Hi Shane,

The response to a DNS query for NS records will need to fit into a udp
packet which limits you to 512 octets. [rfc1035]  The root servers have
the same constraint and cannot exceed 13, so it would be highly unlikely
for you to need more then 2 digits I imagine.

Paulg.
______________________________________________________________________

For PGP Key ID B49E3514,                        mailto:paulg at apnic.net     
send mail with Subject: pgp-key-request        phoneto:+61-7-3367-0490     
Paul Gampe - Technical Manager - APNIC Pty Ltd   faxto:+61-7-3367-0482
______________________________________________________________________