On 04/06/2012 06:23 PM, zxq9 wrote:
On 04/07/2012 03:28 AM, Konstantin Olchanski wrote:
I tend to think that these days one should go back to static IP addresses
for server-type machines, after all, all DHCP, network manager& co do is
assign
the same IP address to the same machine over and over and over again with the
only
variation when they fail to do the boring thing and you have a machine down,
staying
down until somebody physically walks to it to reboot it.
Do you mean there are serious networks that use DHCP by default for
systems other than transient network guests residing in their own little
subnet (like laptops)? And server IP assignment by DHCP... I can't
believe this is really done, or am I being naive about naivete?
It is really being done. I do it for 150+ Linux workstations at work across
several subnets. For 10 systems I might not care, but manual host-specific
configurations are time-consuming to manage. 95% of my boxes are identical
(that itself is a challenge), replicable from bare metal to deployment (via
kickstart and lots of scripting).
That just sounds like a recipe for disaster for a lot of reasons.
Without some thought and preparation any network setup is doomed to get
wacky after a while, and maybe I'm just being too old school -- but
being explicit about setup I've never had a single network problem like
the ones described here, whether letting NM run the show or using the
older networking subsystem.
IMHO, there is no better way than to use DHCP for centralized administration of
all the network parameters. We distribute: IP (statically assigned based on
Ethernet), NIS servers, NIS domain, default router, netmask, netbios servers,
PXE boot params, DNS servers, NTP servers, and hostname.
When we changed our central DNS servers, I didn't have to change every host, I
just changed my DHCP server config file and let the clients pick up the change
when they renewed the lease. Seamless and easy. If I need to change a host or
IP, I just edit the dhcpd.conf file and reboot...great for prepping new
deployments to replace existing systems.
That said, I have seen a few bugs with NetworkManager on EL6 where we've had a
switch go down (bad UPS battery) and a couple clients have gone offline, where
they should have tried to bring the net back up. The few Fedora boxes and EL5
systems running NM have not had these issues. But those are BUGS that should
be replicated, debugged and fixed, otherwise the world won't improve.
And we don't allow any 'guests' on our networks; we never hand out dynamic
addresses for accountability and policy-based reasons.
Just thought I'd toss out another perspective -- it works for us quite well,
and surely there are better methods we could apply too, but that's for another
day.
Regards,
chris
