DHCPD performance -- solid-state disks?
jantman at oit.rutgers.edu
Fri Sep 9 16:38:01 UTC 2011
Well if the server loses power... we've got bigger issues. All of our
wireless controllers are in the same rack, everything is dual PS coming
off of separate PDUs fed from separate UPSes. With generator power.
That's a contingency that we don't really plan for, right or not.
I didn't mention, we also have a script that copies the leases file
every 2 minutes to HDD, and SCPs it to our backup DHCP server (manual
failover). I have yet to see 4.2.0 die unexpectedly, even with ~50+
ACKs/second for months at a time. Then again, we also have a fair amount
of fixed-address hosts, so we restart a few times a day to every few
days. Honestly... if we lose 2 minutes of leases, nobody is going to
complain. Wireless (at least when we're talking about a campus with over
1000 APs, and no policy against unauthorized APs/routers) is still
"best-effort" enough that a minor hiccup for a few users isn't a big deal.
So if, for some strange reason, the box itself loses both PSes at once
or otherwise totally crashes (we're running new HP DL360G6 hardware...
quite reliable IMHO), we lose <= 2 minutes of leases. But have to
manually failover to the backup. If we take a serious power hit that
brings down both circuits, *everything* goes dark, and DHCP no longer
sthaug at nethelp.no wrote:
>> We moved everything (config file, log file, leases file) to a 500MB
>> ramdisk. Our init script is modified so that it persists ramdisk
>> contents to HDD on start/stop/restart, and we also modified our
>> system-wide init scripts to do the same thing.
> What do you do if the dhcpd process dies? Or the server loses power?
> I've always been uncomfortable with the idea of basing everything on
> ramdisk. The fsync in dhcpd is there for a reason...
> Steinar Haug, Nethelp consulting, sthaug at nethelp.no
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the dhcp-users