I've been reading up on some (rather old) mails on this list and I can
see that there have been a few discussions about performance. I thought
I would post what we have at my current job, just as an example for
future reference.
We have two sets of DHCP servers, one set serving guests and the other
set serving everything else. Each set have a standard failover config.
Extra info follow below the stats.
Stats for the set servicing everything but guests:
ISC DHCP version 4.4.1
Number of range statements: 3048
Total number of addresses in range statements: 1641487
Leases in lease file (touched leases): 959816 (58%)
Primary server:
Server type: Physical
Processors: 2 x Intel Xeon E5-2640 @2.40 GHz (20 cores)
RAM: 16 GB
Load average (24h): 0.6
Lease file storage: RAM-disk with periodic backup to HDD
DISCOVERS per day: 127732 (1621 unique clients)
Restart time: 21.2 seconds
Secondary server:
Server type: Virtual
Processors (vCPUs): 4 x Intel Xeon E7-4880 v2 @2.5GHz
RAM: 4 GB
Load average (24h): 0.3
Lease file storage: SSD based virtual disk
DISCOVERS per day: 119340 (1632 unique clients)
Restart time: 16.0 seconds
Stats for the set servicing guests:
ISC DHCP version 4.3.6b1
Number of range statements: 516
Total number of addresses in range statements: 236146
Leases in lease file (touched leases): 194592 (82%)
Primary server:
Server type: Virtual
Processors (vCPUs): 4 x Intel Xeon Gold 6154 @3.00 GHz
RAM: 4 GB
Load average (24h): 0.5
Restart time: 6.5 seconds
DISCOVERS per day: 98265 (4564 unique clients)
Lease file storage: SSD based virtual disk
Secondary server:
Server type: Virtual
Processors (vCPUs): 4 x Intel Xeon E7-4880 v2 @2.50 GHz
RAM: 4 GB
Load average (24h): 0.1
Restart time: 5.7 seconds
DISCOVERS per day: 119340 (4627 unique clients)
Lease file storage: SSD based virtual disk
Restart times are measured with "time systemctl restart dhcpd". I have
only measured this once on every server for these numbers, but it's my
impression that it's nominal.
I have included the number of DHCP DISCOVER messages logged and number
of unique clients as a measure of their approximate load. Data was for
what I hope is a "typical" day, Tuesday December 10th 2019.
There are always at least two DHCP relays and often four, multiplying
the number of DISCOVER messages logged compared to actual client
DISCOVERs. Default lease times are typically 8 days with a few
exceptions.
We have a bunch of misbehaving clients that result in a lot more
DISCOVERs that unique clients per day even though the lease time is
much longer. (We have top-talkers warnings and send requests for
inspection to local people when one client sticks out.)
All the servers also act as a caching resolvers (BIND + anycast) for
all our clients and record (via a rotating "tshark") at least the last
24 hours of all DNS and DHCP traffic for troubleshooting purposes. This
adds a bit to the load average but not much.
We normally restart the secondary server immediately after the primary
and have never seen any problems with this. After a normal restart they
enter functional state with no pause.
--
Peter
_______________________________________________
dhcp-users mailing list
[hidden email]
https://lists.isc.org/mailman/listinfo/dhcp-users
Locked
