optimal dns RR's for 30 servers ?
Kevin Darcy
kcd at daimlerchrysler.com
Sat Sep 11 03:16:54 UTC 2004
elia Mazzawi wrote:
>we have 30 servers in 4 datacenters
>
>3 in A
>12 in B
>5 in C
>10 in D
>
>currently there arranged in 10 groups of 3's but we want to arrange
>them so that we can maximise 3 objectives
>
>1. we have RR redundancy by using different datacenters in each RR
>2. if one server goes down, we will stop sending traffic to the RR's
>its in but still have the load balanced well enough.
>3. we can have up to 900 RR's but less is better.
>
>there are so many arrangements that work, how can you prove this is
>the best arrangement?
>
>i'm working on this too i will give my answer tomorrow
>
This is really a math problem rather than a DNS problem, and therefore
arguably off-topic. As a math problem, you haven't described your
problem set with nearly enough specificity. For instance, what precisely
do you mean by "maximise"?, are there lower or upper bounds on the
number of different datacenters to be represented in each RR?, what
exactly does "load balanced well enough" mean?.
Also, I don't know where you get 900 RR's. According to my calculations,
there are 1,073,736,667 possible RR's consisting of servers from
multiple datacenters (that would be 1,073,741,823 possible non-empty RRs
that can be made from 30 servers, minus all of the non-empty RRs which
consist entirely of servers from a single datacenter -- 7 A-only RR's,
4095 B-only RR's, 31 C-only RR's, and 1023 D-only RR's).
- Kevin
More information about the bind-users
mailing list