full socket buffers

Miroslav Lauš mirek.laus at gmail.com
Sat Dec 18 19:24:21 UTC 2010


we're running ISC DHCP 4.1.1p1 on FreeBSD 7.3-RELEASE box
for about 30k hosts with various leasetime from 4 hours to 7 days.
Time to time we experience significant performance loss while
clients are unable to get a lease. In the time the UDP full socket
buffer is constantly growing:

FreeBSD 7.3-RELEASE-p2 (GENERIC) #0: Mon Jul 12 19:04:04 UTC 2010
[root at dhcp1 ~]# netstat -s -p udp
	148350844 datagrams received
	0 with incomplete header
	0 with bad data length field
	510 with bad checksum
	1257964 with no checksum
	369989 dropped due to no socket
	0 broadcast/multicast datagrams undelivered
	6180438 dropped due to full socket buffers
	0 not for hashed pcb
	141799907 delivered
	442940589 datagrams output
	0 times multicast source filter matched

The only solution to this situation is to restart the dhcpd process.
We've change the sysctl variable:

kern.ipc.maxsockbuf: 8388608

and to even higher value, but no luck. The issue comes in about weekly period.

How could I tell the cause of the full buffer socket?


More information about the dhcp-users mailing list