[KLUG Members] MAC/IP conflict

[Adam Williams]

>First off, thanks for talking me down. :) I had two 4am nights in a
>row and wasn't feeling, ah, well, human for a while there.
>Since the network isn't failing right now, I'm going to install the SCSI
>card and DAT and do a backup, as we haven't had one since Wednesday. I'm
>also curious to see what a reboot will do to our environment.

I'd boot the server into memtest overnight and see if it comes up with
any errors.  If the RAM isn't ECC.  Problems with cheap-o-RAM can occur
suddenly.

What do the ifconfig packet counters look like?  Are they accumulating
at an abnormal rate?

>>>Because the source of the conflict is far from clear.
>>Are you sure it's a IP conflict?  I thought you confirmed that, and I
>>was going on that assumption.
>With no logs giving errors, but a couple WinXP Pro boxes giving DHCP/IP#
>conflict errors, I thought that one of the two other businesses in this
>building might have patched into our hub by accident and two boxes were
>duking it out over our server's IP#. However, a thorough trace of all
>our wiring has removed that possibility (but left us an updated map!).

Ah ha!  What does your leases file look like?  Possibly corrupted?  When
did they acquire their conflicting lease?  Possibly a LinkSys card with
a wandering MAC?

>We have several laptop users, but only one left her laptop here for the
>weekend.
>Two possibilities exist: that the IP conflict reported by the WinXP
>boxes were caused by the server being unavailable to service the DHCP
>request, or the server was knocked off by a misconfigured box ignoring
>DHCP and trying to seize a taken IP#. However, after a thorough shutdown
>of all the machines on the network, then bringing them up one by one,
>I've not seen this problem since. With the single DHCP server getting
>knocked off the net for minutes at a time, the chances of DHCP remaining
>coherent were not good.

Mm.  I'd disagree.  I've taken DHCP servers off networks for *HOURS* at
a time,  the DHCP protocol takes into account intermittant
unavailability during release & renew operations.  It is actually really
hard to make ISC DHCPd incoherent.  I've never had it happen except when
the filesystem holding the lease database got trashed.

Did you stop dhcp service and dust the leases database before booting
everything back up?