dns RR method is not equal balanced?

[John Wobus]

On Mar 29, 2011, at 10:49 AM, Tony Finch wrote:
> Kay <chokh at daumcorp.com> wrote:
>>
>> some domain has 12 IPs but traffic of the server is not equal.
>> The traffic of 11 IPs is same and just 1 IP is higher than others.
>
> If you use round-robin DNS you are relying on the clients not to muck
> around with the responses they get from your DNS server. If they sort
> them, for example, that will mess up the balancing. For example RFC  
> 3484
> screws it up.


In theory, if the glsb can be configured to act the way bind does,
you could isolate the problem to client behavior.  Obviously
bind's round-robin behavior could be checked "in the lab"
with a test name.

Seems unlikely given your description, but if the clients vary
as to how much load they impose for each DNS lookup they do,
you're more likely to see imbalances.  For some clients, TTLs
might contribute to this.  However, given enough time, the
pattern should shift, i.e., it wouldn't always be that this
one IP gets most of the load.  If the clients are daemons
that stick to a server for months based upon a single
DNS lookup, then this time might be very long.

If you're dealing with typical web hits, such a scenario
is unlikely.

John