Memory footprint of dhcpd process steadily increasing

[Oscar Ricardo Silva]

Two of our dhcp servers failed with "out of memory" messages:

   Mar 30 07:39:17 DHCP-SERVER out of memory [8971]
   Mar 30 07:39:19 DHCP-SERVER last message repeated 42 times

(two hours later the second server in the active/active failover 
relationship had the same problem)

It seems the dhcpd process' memory footprint is steadily increasing 
until the system runs out of memory and dhcpd dies.  We do use OMAPI but 
we run version 3.1.1 which fixes that related memory leak.  I found one 
other mention of this problem posted by Steinar Haug:

   <https://lists.isc.org/pipermail/dhcp-users/2008-March/005951.html>


Restarting the process returns memory usage back to normal.  Here I 
restarted the process at 13:23 :

Time		RSS	VSZ	COMMAND		%MEM
2009-03-31.1132	330724	331980	/usr/sbin/dhcpd	15.9
2009-03-31.1134	331604	332928	/usr/sbin/dhcpd	15.9
2009-03-31.1136	332524	333880	/usr/sbin/dhcpd	16.0
...
2009-03-31.1316	384248	385540	/usr/sbin/dhcpd	18.5
2009-03-31.1318	385164	386500	/usr/sbin/dhcpd	18.5
2009-03-31.1320	386748	388044	/usr/sbin/dhcpd	18.6
2009-03-31.1322	388008	389348	/usr/sbin/dhcpd	18.6
2009-03-31.1324	131352	132872	/usr/sbin/dhcpd	6.3
2009-03-31.1326	132192	133728	/usr/sbin/dhcpd	6.3
2009-03-31.1328	133296	134804	/usr/sbin/dhcpd	6.4
2009-03-31.1330	134584	135956	/usr/sbin/dhcpd	6.4
2009-03-31.1332	135504	136792	/usr/sbin/dhcpd	6.5


Is this normal behavior?  Is this known but abnormal behavior?

Both servers are running Red Hat Enterprise Linux Server 5.3 with 2GB of 
memory and acting in an active/active failover relationship, servicing 
approximately 20,000 leases.

Any information would be appreciated.


Oscar