dns server and client load blancing behavior

Brad Knowles brad.knowles at skynet.be
Fri Sep 14 22:54:39 UTC 2001


At 10:45 AM +0800 9/14/01, Æ}Ə©[ wrote:

>  if I has following RR
>  ms1    IN    MX    0    ms1_1
>             IN    MX    0    ms1_2

	First off, don't use underscores.  They are illegal in hostnames.

>  ms1_1    IN    A    10.10.10.1
>                 IN    A    10.10.10.2
>                 IN    A    10.10.10.3
>  ms1_2    IN    A    10.10.10.4
>                 IN    A    10.10.10.5
>                 IN    A    10.10.10.6
>
>  then query mx Record for ms1,will name server response all the above RRs for
>  client,and the what the RR order?

	The order will be non-determinate, in accordance with the RFCs. 
Some servers will do a round-robin (listing one first on one query 
and then the other first on the next query), and other servers will 
use other randomization mechanisms.

>  How long do Mircrosoft client or any other client will cache the RR?

	Unfortunately, Microsoft OSes will only ever attempt to contact 
the first IP address listed for any machine that they look up (they 
simply never see the second or other IP addresses), and they will 
cache this one IP address forever -- well, until they get rebooted.

	So, the generally accepted solution is to have an 
application-layer proxy, or a layer 4 load-balancing switch sitting 
in front of however many back-end machines you have, and then you 
list only the one IP address of the application proxy or L4 
load-balancing switch, and it hides the rest of that complexity from 
the Microsoft client.

-- 
Brad Knowles, <brad.knowles at skynet.be>

H4sICIFgXzsCA2RtYS1zaWcAPVHLbsMwDDvXX0H0kkvbfxiwVw8FCmzAzqqj1F4dy7CdBfn7
Kc6wmyGRFEnvvxiWQoCvqI7RSWTcfGXQNqCUAnfIU+AT8OZ/GCNjRVlH0bKpguJkxiITZqes
MxwpSucyDJzXxQEUe/ihgXqJXUXwD9ajB6NHonLmNrUSK9nacHQnH097szO74xFXqtlbT3il
wMsBz5cnfCR5cEmci0Rj9u/jqBbPeES1I4PeFBXPUIT1XDSOuutFXylzrQvGyboWstCoQZyP
dxX4dLx0eauFe1x9puhoi0Ao1omEJo+BZ6XLVNaVpWiKekxN0VK2VMpmAy+Bk7ZV4SO+p1L/
uErNRS/qH2iFU+iNOtbcmVt9N16lfF7tLv9FXNj8AiyNcOi1AQAA


More information about the bind-users mailing list