On Mar 5, 2012, at 15:01 , Stephen Berg (Contractor) wrote:

> On 03/05/2012 07:37 AM, Stephen Berg (Contractor) wrote:
>> I'm testing a way to install/upgrade some remote systems.  What I'm doing is hand jamming a change in /boot/grub/grub.conf to point to /boot/vmlinuz and /boot/initrd.img from the /image/pxeboot directory off the install DVD, tried both 6.1 and 6.2.  Both files have been copied to /boot on the test server.
>> 
>> When I boot to this kernel/image combination I use the following boot parameters to get a psuedo netboot/pxeboot installation started:
>> 
>> ks=nfs:<IPADDRESS>:<PATH to kickstart> ksdevice=link vnc vncpassword=<PW>
>> 
>> The problem I'm running into is that the ksdevice parameter seems to be getting ignored.  I've tried ksdevice=ethX with the appropriate network interface name, ksdevice=link and ksdevice=<MACADDR> but the system consistently stops at the screen asking me to choose which interface to use.
>> 
>> I tried a similar method using a Fedora 15 vmlinuz/initrd.img and ksdevice=link and it worked as expected.  Is there some magic I'm missing or is that ksdevice parameter not functioning properly from the install DVD's copies of vmlinuz/initrd.img?
>> 
>> I'd really like to get this working to help with some remote system installations.
>> 
> 
> Forgot to mention this system has two NIC's, both enabled but only one has an active link.  In SL6.x the active link shows up as eth1, during the test with Fedora 15 that I mentioned the active link came up as eth0.

ksdevice=link seems to work for us when reinstalling SL5 systems with SL6. But maybe that's just because we have installed all systems with pci=bfsort since the SL4 times, making device numbering consistent across kernel versions (and we use only eth0 for PXE). Using pci=[no]bfsort, it should always be possible to make the interface in use primary one. But then I'm not sure what Networkmanager or the new parallel udev device discovery may be doing...

And since 6.1, some (recent Dell) servers have biosdevname enabled by default. Which makes the first onboard interface "em1", for example :-) The advantage is that this will always match the physical socket labelled "1" on the server.

> I can get to this system to test pretty easily, the systems I want to do this on if I can fix this issue I can't.  No iDrac type interface, I can remote in and reboot them, but there's no easy way to select a network boot without physically going to the machine and hooking up a keyboard/monitor.  And that's what I'm trying to avoid if possible.

-- 
Stephan Wiesand
DESY - DV -
Platanenallee 6
15738 Zeuthen, Germany