Manipulating the Round robin

Kevin Darcy kcd at chrysler.com
Tue Jan 15 00:07:36 UTC 2008


Radi Tzvetkov wrote:
> Hello list,
> I am trying to find more information on how to manipulate the round robin.
> I have a website.com that responds with 1.1.1.1 and 2.2.2.2 and these are
> load balanced via round robin.
> is there any way to manipulate the round robin and have the web site return
> the IPs in the same order every time?
>
> say disable the round robin?
>
>   
Yes, there is (see rrset-order and/or sortlist), but before you start 
down that path, realize that *any*other*resolver* that caches the 
information is going to employ whatever sorting algorithm it deems 
appropriate, when answering from its cache, which in many cases reduces 
to round-robin as the Lowest Common Denominator. Manipulating the order 
of address records only works reliably, therefore, when you control 
*all* the resolvers between the client and the authoritative 
nameservers. This is not feasible for Internet names, but has limited 
applicability to some intranets where the resolvers are all under 
central control.

You can defeat caching somewhat by lowering your TTLs to anti-social 
levels, but do you really want to be that rude and impolite, not to 
mention squandering your own nameserver resources?

My other question would be: what are you trying to achieve by having the 
addresses be given out in the same order every time? Do you have one 
"strong" server and one "weak" server? How then is the "weak" server 
going to handle the load if the "strong" server fails? Generally 
speaking, it's a better strategy to beef up both servers to 
approximately the same capacity, and then use simple round-robin, or a 
commercially-available load-balancing solution, to split up the load 
approximately equally between them. If one of the servers fails, at 
least then you have a fighting chance of the other server being able to 
handle the load.

                                                                         
                                    - Kevin




More information about the bind-users mailing list