Cisco Distributed Director
Kevin Darcy
kcd at daimlerchrysler.com
Fri Jun 16 22:05:45 UTC 2000
I still don't get it. BIND 4.9.7 implemented *true* round-robin, so the load
should be split as equally as possible for DNS-based load-balancing. Could
it be that the low TTL is *causing* some sort of skewing? I seem to recall
that BIND 4 didn't deal very well with 0 TTL's.
Have you tried it with a *higher* TTL value? I know it seems rather
paradoxical...
Regardless, you should upgrade.
- Kevin
Dan Considine wrote:
> Here's the environment.
>
> Primary DNS server (Bind 4.9.7 / Solaris 2.6)
> Secondary DNS server (Bind 4.9.7 / Solaris 2.6)
>
> Distributed Director (Cisco IOS 11.1.(20)IA)
>
> I have delegated dd.us.ml.com from my DNS servers to
> the DD
>
> dd.us.ml.com IN NS wfddd1.ml.com
> IN NS huddd1.ml.com
>
> On the DD I have a director host setup with 8 ip's with
> equal portion metrics, TTL = zero
>
> wfdbbsocks-p.dd.us.ml.com (8 ip addresses)
>
> Clients have the following:
>
> Primary and Secondary DNS servers setup as their resolvers.
>
> Start Openbloomberg, config file states wfdbbsocks-p.dd.us.ml.com
> as it's socks server.
>
> Client tries to resolve socks server by way of resolvers, resolvers
> go to the DD (wfddd1 or huddd1) to get an A record, DD returns an ip
> by way of equal portion metrics (round robin), resolver returns ip
> to client and caches this A record with a TTL of zero. Now, this is
> fine on a client by client query, works fine. The problem is when we
> restart our proxy servers. We have 3,000 users connected over the 8
> machines, once the client loses it's connection it tries to get another
> ip. So, we have 3,000 queries sent to DNS, DNS starts handing out the
> last cached ip while it tries to get another 1 from the DD, by this time
> it's too late, maybe 500 clients already received the same ip which
> defeats
> the round robin setup. I need to find a way to have our DNS servers send
> the client to the DD for the A record (non-recursive dns).
>
> Dan Considine wrote:
>
> > Is there a way to setup non-recursion on an NS record in Bind 4.9.7? I
> > need to have client queries sent to my Distributed Director for
> > resolution, TTL on the Director is set to "0", root server takes 1
> > second to cache this record which
> > defeats the purpose of the "0" TTL. Any ideas?
> >
> > Thanks in advance,
> >
> > Dan.
More information about the bind-users
mailing list