Named resulting in OOM condition

[Hans Kristian Rosbach]

We are running a simple caching named process on a separate
computer here. Every two weeks it needs to be rebooted due
to running out of memory. It can survive for a while further
on swap, but that slows things down terribly.

After restarting named the memory is still in use according
to 'free':
             total       used       free     shared    buffers    
cached
Mem:        775664     760064      15600          0        736     
11884
-/+ buffers/cache:     747444      28220
Swap:      1566296       9108    1557188

But according to ps all processes use a 0.0-0.6% ram. (ps -Ae vx)

from slabinfo I got the following interesting lines:
biovec-(256)         256    256   3072    2    2 : tunables   24   12   
0 : slabdata    128    128      0
biovec-128           256    260   1536    5    2 : tunables   24   12   
0 : slabdata     52     52      0
biovec-64            256    260    768    5    1 : tunables   54   27   
0 : slabdata     52     52      0
biovec-16            256    260    192   20    1 : tunables  120   60   
0 : slabdata     13     13      0
biovec-4             256    305     64   61    1 : tunables  120   60   
0 : slabdata      5      5      0
biovec-1          6341638 6342238     16  226    1 : tunables  120  
60    0 : slabdata  28063  28063      0
bio               6341638 6341798     96   41    1 : tunables  120  
60    0 : slabdata 154678 154678      0

Seems like bio is taking up nearly all the memory, what can cause this?
Any way to force it to go away? =)

This problem has been there for a long time, we have this problem on
all our nameservers but always thought it was a bug due to us reloading
a huge config every half hour. 2GB ram makes those boxes run for a few
months before they need a reboot.

But this caching nameserver should not be such a special case, and since
it caches all our RBL queries it seems to run out of memory much faster.

This problem has followed us atleast since FC2-test2, and now FC3.
I just upgraded to bind-9.2.5-1, but I seriously doubt that it will have
any significant effect.

Ideas?

-HK