redundant servers

Kevin Darcy kcd at daimlerchrysler.com
Wed Nov 19 00:10:10 UTC 2003


RYAN vAN GINNEKEN wrote:

>Ok here is my scenario i have 3 registered domains and two Freebsd boxes 
>running bind 9 postfix 2 and apache 2.  These servers are on separate 
>networks server one is master for itself and slave for the remaining 2 
>domains on the server 2.  Server 2 is slave for server 1 and master for 
>a couple of its own domains yet only has one ip address.  This i seem to 
>have figured out over the past couple of years what is confusing me 
>right now is how do i get the servers to back each other up i mean  if 
>one goes down  the other one takes over dns web page mail sever 
>everything.  seems like a great concept but how do i achieve this i 
>think i got the dns stuff  but what about the rest dns seems to be 
>everything in these situations so  i hope this is not to far off topic
>
Nameserver failover is automatic, as long as both servers are published 
in the NS records of the zone and the delegation records.

Mail failover is also automatic with the use of MX records. For each 
domain, you could have one of the servers be primary, and the other 
failover, or you could (roughly) load-balance the mail traffic between 
the servers. It all depends on how you set the "preference" fields on 
the MX records.

For web failover, DNS is not a very good solution, because there is no 
web equivalent of MX records, and no web clients yet understand SRV 
records, which can be thought of as bringing MX-record-like failover and 
loadbalancing capabilities to any record type. Even if SRV records were 
understood by browsers, there would still be a timeout delay for 
failover, which is not so critical in a store-and-forward system like 
email, but *very* noticeable when a user is sitting and waiting for a 
page or an image to load. Using DNS, you could assign the addresses of 
both webservers to your website names, but you lose control of the 
ordering of those records once they get cached by other nameservers on 
the Internet, so you're still faced with the possibility of long 
failover delays when one of the webservers is down. Another DNS-based 
approach is to automatically change the A record for one of more 
websites when the primary webserver goes down. However, caching gets in 
the way here too, and you'd have to reduce the TTL (Time-to-Live) values 
on your records to anti-social values in order for this "switcheroo" to 
be dynamic enough for an average web user.

For this reason, many folks pay a lot of money for fancy standalone 
load-balancing devices from Cisco and other vendors to enhance the 
performance and/or availability of their websites...

                                                                         
                                 - Kevin




More information about the bind-users mailing list