notworking limits on 8.2.3R

[Kevin Darcy]

This is standard behavior for BIND 8 when it reaches the configured memory
limit -- crash and burn. Supposedly BIND 9 does (or at least will) deal with
out-of-memory conditions more gracefully.

But, I'm curious why you think there is a strong correlation between the
number of clients and the amount of memory that named uses (?) Certainly
more clients will jack up your CPU usage and network traffic, and this might
have an *indirect* effect on (mostly kernel) memory usage, but predominantly
named uses memory for cached answers, and this is even less directly
related, if at all, to the number of clients it is serving -- if all of the
clients are asking the same questions, then the cache size won't grow much,
no matter how many clients you add. Only if your clients are making widely
divergent queries is there likely to be a significant correlation between
number of clients and cache memory usage.

If you're *already* at your memory limit, then it's time to think about
adding more memory, moving DNS to a dedicated box (if it's currently sharing
resources with other functions), or spreading the DNS query load between
more machines (although for similar reasons as discussed above, this is only
likely to gain *modest* memory savings).


- Kevin

cjm at Enemy.ORG wrote:

> Hello,
>
> I'm having a problem with bind 8.2.3R and setting limits.
> i.e. when i limit datasize to 48M and the named-process reachs
> this limit it crashes. OS is netbsd1.3.2 - could it be a OS-related
> problem with process limits? Im having a increasing number of
> clients so i need to set a limit.
>
> thanks,
> cjm
> --
> ** PGP-Key-ID: 0x975B2F19 - CJM17(-RIPE),SIL-MNT     http://www.enemy.org
> ** --  "You can tune a file system, but you can't tune a fish." tunefs(8)
> *************************************************************************