So I finally broke down and installed RH 9.0 on the head node, then installed Oscar 3.0.
All of the machines are Dell PowerEdge 1750s with Broadcom Gb NICs. I used Frank Crawford's patch kernel, initrd.img and boel_binaries.tar.gz files, which I placed in /usr/share/systemimager/boot/i386/standard/ so that there would be drivers for the NICs.
The imaging proceeded successfully, with none of the silliness which accompanied it on Man 9.2 with "error loading module 1" and the like. I can image a node successfully, and that node boots up into RH9.0.


However, the kernel (I selected the default one: 2.4.20-8smp was what was installed on the nodes) now has the same old problem of not being able to detect my NICs! At least, that's what I've self-diagnosed as the problem, since it has no network capability, I didn't see a driver load in dmesg, and ifconfig -a shows only loopback. Well, keep reading. :)

Since it had been mentioned that this has been made to work with RH9.0 on PowerEdge 1750s, I was hoping someone could give me a cookie cutter solution. If not, could anyone suggest the BEST way of imaging the nodes with OSCAR/SIS to include the driver (bcm5700, I believe)? Is it to statically link it into a kernel and use that? (I can't imagine this, since every other thing works with modules.) Keep reading. :)

The /etc/modules.conf file on my head node is now:
alias eth0 tg3
alias eth1 tg3
alias scsi_hostadapter megaraid
alias usb-controller usb-ohci

The /etc/modules.conf file on a cluster node is:

# BEGIN: Lines added by System Configurator
alias scsi_hostadapter mptscsih
alias scsi_hostadapter1 mptscsih
# END: Lines added by System Configurator

If I edit it to look like

alias eth0 tg3
alias eth1 tg3
# BEGIN: Lines added by System Configurator
alias scsi_hostadapter mptscsih
alias scsi_hostadapter1 mptscsih
# END: Lines added by System Configurator

The node that I make that correction on now has net access, and can ping the head node, the head node can ssh into it with no password prompt, and it seems to generally function correctly. However, manually adjusting the /etc/modules.conf file on every node by console is an ugly solution at best--I have to log in to each node manually, edit a file, save it, and reboot. If I ever want to reimage or use different images, this won't be practical, and there won't be any way to automate it because there's no net access to the boxes. There must be a good way to do it within SIS/OSCAR. I thought of changing the /var/lib/systemimager/images/oscarimagerh/etc/modules.conf file, but such a file does not exist. What's the best way to fix this, guys?


Also, just as an aside, on the node which I fixed thusly, it wouldn't correctly mount /home which is being exported from the head node, giving me the error:


Dec 15 15:46:32 elf4 kernel: tg3: eth0: Link is up at 1000 Mbps, full duplex.
Dec 15 15:46:32 elf4 kernel: tg3: eth0: Flow control is off for TX and off for RX.
Dec 15 15:46:32 elf4 kernel: ip_conntrack version 2.1 (8192 buckets, 65536 max) - 292 byt
es per conntrack
Dec 15 15:46:38 elf4 mount: mount: RPC: Remote system error - No route to host


If one subsequently logs in, a simple

mount /home

works just fine, and /home mounts correctly (implying that the fstab entry is correct, the exporting is correct from the head node, and the /etc/hosts file on the node is correct as well.) This certainly is small potatoes--I could just add a little S99 script on each node which did nothing but mount -a. However, I find this a little strange that it's not automatically being mounted during start up, but then mounts perfectly afterwards. The other problem is the bigger one, but I was hoping someone had seen something like this in RH9.0 and had fixed it.

Thanks in advance for your help,

Jason
--------------
Jason Hlady, B. Sc., M. Sc. (Chem), Adv. Cert. (Comp. Sci.)
Programmer/Analyst (Bioinformatics Specialist)
U of Saskatchewan, Bioinformatics Research Laboratory (BIRL)
[EMAIL PROTECTED] (306) 966-2075



-------------------------------------------------------
This SF.net email is sponsored by: SF.net Giveback Program.
Does SourceForge.net help you be more productive?  Does it
help you create better code?  SHARE THE LOVE, and help us help
YOU!  Click Here: http://sourceforge.net/donate/
_______________________________________________
Oscar-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/oscar-users

Reply via email to