Arnaud Girsch writes:
>>>> How are people out there implementing round-robin DNS? I have "n"
>>>> servers acting as www.wherever.com and I want to split the traffic
>>>> bewteen them. Of course, if I could do a proper load balancing that
>>>> would be great, but a simple "n" way split is a big help already.
Firstly an answer... you can use bind-4.9.3 if you compile it with
ROUND_ROBIN defined. I actually modified bind-4.9.3beta17 so that a
round_robin=N directive in named.boot could be used to turn it on.
The N is the max number of addresses to return - several PC TCP stacks
corrupt answers that contain more than 4 addresses. I delegate a
sub-domain rr to a couple of round-robin servers which play this game
with TTL=0, and the normal servers contain cname records for say www
which point at www.rr.
For real load balancing, I like the perl lbnamed which was in a LISA
paper a couple of years ago. It uses a client/server deal to
determine which serevrs are least loaded... I did a new module for it
so that you could give it a fixed config that would deal out A 90% of
the time and B and C 5% each. This suited my needs well.
>'just show me some DNS implementation that doesn't round robin .. ?
bind-4.9.3 (well beta17 at least) does not _unless_ you define
ROUND_ROBIN.
>Probably few of them are not round robin'ing, but I have to disagree when you
>say that "any off-the-shelf" doesn't.
Just about every DNS should be running bind-4.9.3 and without
ROUND_ROBIN it _will_ return the address list the same every
time. I've tested it - I have sites that rely on it.
--sjg
--
Simon J. Gerraty <sjg @
zen .
void .
oz .
au>
#include <disclaimer> /* imagine something _very_ witty here */
References:
|
|
