bind-users Digest, Vol 979, Issue 3

TMK engtmk at gmail.com
Fri Sep 23 09:45:42 UTC 2011


On Sep 23, 2011 3:06 AM, "Chris Buxton" <chris.p.buxton at gmail.com> wrote:
>
> On Sep 22, 2011, at 3:50 PM, TMK wrote:
> On 9/21/2011 5:00 PM, TMK wrote:
> >>> I have couple of questions.
> >>>
> >>> bind cache memory limit is 4GB. can I increase it. or this is
hard-coded limit.
> >>>
> >>> i'm running the x64 bit version.
> >> You can _try_ to raise that limit above 4Gb (see the various
> >> configuration elements under "Operating System Resource Limits" in the
> >> ARM), but your OS may still limit you. You'd need to look at your OS
> >> documentation to see how to raise "soft" and/or "hard" limits.
> >
> >
> > if i try to set the max-cache size to more that 4GB it says very large
number.
> >
> > and the default if left un-set it should be unlimited as per ARM.
>
> It may be that named can't accept a size larger than 4GB. Any input from
ISC? Mark?
>
> >>> also to increase the cache hit ratio I have created script to query my
> >>> dns for the top 1 million sites. would this give any performance
> >>> advantages or is it useless.
> >>>
> >> Since a large percentage of those top sites use DNS-based
load-balancing
> >> with small TTLs on their records, it's pretty useless.
> >>
> >> Of course, you can measure your cache hit ratio to see if it's really
> >> helping you or not. Real data always beats pure speculation.
> >>
> >>
> >>                                                                     -
Kevin
> >
> > to use more memory I have created two instances of the bind. one is
> > listening on the machine IP addr. and the other on the loop-back
> > address (127.0.0.1).
> >
> > i have set the first instance (which is accepting the client
> > connection) to forward only and the forwarders is the 127.0.0.1.
> >
> > would this give me better cache hit ratio.
>
> Probably not. There will be differences between them, in that the server
on the loopback address will end up caching stuff about parent domains and
name servers, which the client-interacting service won't see. But you'll be
wasting a lot of memory with duplicated data.
>
I totally agree but the strange thing we noticed response time improvements
for the un-cqched quiries for the DNS. When the loop back instance is up and
running.

> 4 GB is a lot of space in the cache. Bearing in mind that DNS records
expire from cache when their TTLs run out, if you're seeing the cache run
out of space, you must be fielding a lot of traffic. Would you be better off
using multiple separate boxes?
>

We already have multiple boxes this only one of them.
> Regards,
> Chris Buxton
> BlueCat Networks
On Sep 23, 2011 3:06 AM, "Chris Buxton" <chris.p.buxton at gmail.com> wrote:
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/bind-users/attachments/20110923/0a51e2da/attachment.html>


More information about the bind-users mailing list