dhcp 3.1.3 failover performance questions

K. K. Tam k.k.tam at ust.hk
Thu Mar 18 08:22:20 UTC 2010


We are running DHCP 3.1.3 failover with 2 Intel Xeon Servers (Dual-Core 
5130 2GHz
and 2G RAM) on CentOS 4.7 and syslog on local disk.  The set up is 
serving about
13000 leases and 50% of them with default lease time from 30 minutes to 
2 hours
and the rest is about 2 days.  The normal lease usage ranges from 40% to 
60% of
the total leases and 50% of them are in the relative short lived subnets.

Lately we have some dhcp performance degradation problem after adding 
some short
lived subnets, which is about 1500 leases and included in total leases 
mentioned.
The IP request cycle (dhcpdiscover, dhcpoffer,dhcprequest and dhcpack) 
can be from
10 to 12 seconds in high usage period, which the total lease usage is 
close to
50%.  Normally the IP request cycle can be done in 1 to 3 seconds.  
Based on the
dhcpd log, we observe that the time between dhcpoffer and dhcprequest is 
beyond 6
to 7 seconds in the relatively high usage period and is undesirable.

I have the following questions to seek your advices.

1. Is there anything we can do to improve the performance, say back to 
normal, in
    the direction of hardware resources and/or software configuration?
    Currently the number of messages printed in the dhcpd log file is 
about 20 lines
    per second.

2. For the capacity planning, we might consider to add more leases to 
support in
    future.  Is there any guideline on the hardware resources planning?

3. There is no performance difference between dhcp server with VLAN 
trunk port support
    and with the dhcp relay agent (on the network router).  Is that correct?
    In our environment, we have a couple of VLANs set up and they are 
normal in high
    usage period yet they have relatively few leases to support.

Thanks in advance.


Regards,
K. K. Tam




More information about the dhcp-users mailing list