netdump-server is a server that listens on the network for connections
from kernels using the netconsole/netdump module. When such a machine
crashes it contacts the netdump-server on the specified ip/port. The
server then proceeds to log console messages, including the oops message,
and a kernel memory dump in a subdirectory of /var/crash, and then
reboots the machine.

You should load the module on the client machine by installing the
netdump package, setting the server IP address in the file
/etc/sysconfig/netdump, and then running "service netdump start".
This stores the MAC address of the server; if you expect that to
change regularly, you might want to create a cron job that does
"service netdump reload" to get the new MAC address.

On the server machine, run the netdump server as a user with write
privileges to /var/crash (as packaged, the init script will run it
as the netdump user, and the netdump user owns /var/crash). It will
keep running and writing in /var/crash. You can trigger custom code
by placing scripts called netdump-start, netdump-crash,
netdump-nospace, and netdump-reboot in /var/crash/scripts.

netdump-start is called whenever a remote machine loads the netdump
module, and is given the ip address of the machine as the argument.

netdump-crash and netdump-reboot is called when machine crashes and
then reboots, and are given the ip of the crashed machine and the
directory where logs and memorydumps are written as arguments.

netdump-nospace is called when /var/crash runs out of space.  It
gets one chance to clean up enough space for the dump in question
to proceed.

On the server machine, you may also want to set up syslogd to
receive network traffic (if that is not already enabled).  To do
that, edit /etc/sysconfig/syslog and add the "-r" and possibly
"-x" options to SYSLOGD_OPTIONS, as documented in comments within
that file.  You only need that if you have set up SYSLOGADDR on
the clients.