round robin

WebReactor Networks bind at webreactor.net
Sun Feb 17 14:19:14 UTC 2002


Today I am the fool.  Sorry to post erroneous information based on ancient recollection rather than research, and thank you for the corrections.  Thus I learn...

- - - - - - - - - - - - - - - - - - - - 
From: Alan Sparks <asparks at quris.com>

Does on mine.  BIND 8 (at least) rotates through A (and later revs any)
records.  Problem can be when clients cache and fixate on one of the
addresses...

I have something like this on my name servers:
corpldap  IN A 208.169.17.136
          IN A 208.169.17.250

and dig shown alternate order on each query.

Don't use ping to test, use dig:
    dig @nameserverip host.yourdomain.tld 
Do it a few times, note the order of the returned records.

-Alan

- - - - - - - - - - - - - - - - - - - - 
From: Len Conrad <LConrad at Go2France.com>
To: bind-users at isc.org

>More to the point, BIND does not do any round-robining.

To quote from your URL below:

"This is the current "round robin" function, and it has been used quite 
successfully for several years in sharing load among several machines."

RRset-order default is "cyclic".  dig any RRset 3 or 4 times and see the 
physical order of the records change.

>BIND simply returns both of the A records.

but they both can't arrive exactly simultaneously, so they arrive in a 
physical ordering.

>The client application chooses which of those A records to use.

could be, but vast majority is looking for one A record, and will take the 
first in physical order.

>See < http://www.isc.org/products/BIND/docs/bind-load-bal.html > for more 
>info.

I suggest you see it yourself.

Len

- - - - - - - - - - - - - - - - - - - - 
From: Jim Reid <jim at rfc1035.com>

>>>>> ">" == WebReactor Networks <bind at webreactor.net> writes:

    >> More to the point, BIND does not do any round-robining.  Period.

This is utterly false. Period.

- - - - - - - - - - - - - - - - - - - - 
From: Peter H%E5kanson <phn at icke-reklam.ipsec.nu>

WebReactor Networks <bind at webreactor.net> wrote:
> More to the point, BIND does not do any round-robining.  Period.  BIND simply returns both of the A records.  The client application chooses which of those A records to use.  In your case the "client application" is the Windows resolver.

Close.  BIND by default returns all A records, the order between them is 
randomized. This is controllable with the "rrset-order" statement in bind-9.

Next, what has been observed is that the ping command will do 
one name lookup and then use the first address for all "pings".

Repeating the ping will most likley show a random selection ( unless 
some vendor has created a "nameserver-cache". This cache should in
almost all cases be turned off and disabled)

> See < http://www.isc.org/products/BIND/docs/bind-load-bal.html > for more info.
- - - - - - - - - - - - - - - - - - - - 

WebReactor Networks wrote:
> 
> More to the point, BIND does not do any round-robining.  Period.  BIND simply returns both of the A
records.  The client application chooses which of those A records to use.  In your case the "client
application" is the Windows resolver.
> 
> See < http://www.isc.org/products/BIND/docs/bind-load-bal.html > for more info.
> 
> -------- Original Message --------
> Subject: RE: round robin
> Date: Fri, 15 Feb 2002 11:04:04 -0500
> From: "Haygood, David" <david.haygood at pgnmail.com>
> To: "'bind-users at isc.org'" <bind-users at isc.org>
> 
> If your Windows Desktop is 2K or XP its cache is 24 hours.
> 
> -----Original Message-----
> From: Brian Noecker [mailto:bnoecker at jabber.com]
> Sent: Wednesday, February 13, 2002 6:26 PM
> To: 'bind-users at isc.org'
> Subject: round robin
> 
> Is there a trick for round robin to work correctly?  I've got the following
> scenario:
> 
> Two Bind 9.2.0 jailed servers, one master, one slave with either of the
> following A record options:
> 
> test    10    A    10.2.2.1
> test    10     A    10.2.2.2
> 
> or
> 
> $TTL 2h
> 
> test     10    A    10.2.2.1
>      10    A    10.2.2.2
> 
> Now, on the both servers, they use themselves as primary DNS in resolv.conf.
> The master is on a solaris box, the slave a RH7.2 box.
> 
> When I'm on the master itself and ping test, 10.2.2.1 is the only answer I
> get back.  When I'm on a windows desktop using the master as its DNS server,
> I only get 10.2.2.1.  When I'm on the slave server and ping, I get about a
> 30/70 split on my returns.
> 
> Should I expect consistency on the round robin nature of Bind?
> 
> Thanks in advance.
> 
  


More information about the bind-users mailing list