Bios/Firmware update

[Rich Hickey]

We're starting to get hit by Dell wanting us to do Firmware updates to
resolve memory issues on our Dell/LNXI 1950 clusters. I also have a
couple of nodes that need system boards replaced and we've delayed
this due to the question of what happens.

The question I have is "What magic" was done by LNXI to the firmware
and what will happen when I replace it with Dell standard firmware?
Will the node go into la la land and not boot through clusterworks?
Not sure if that's a bad thing or not really. It would force us off
of the product, which isn't supported anymore.

If anyone has any ideas before I blow away a node testing I'm open to
suggestions.

Thanks.

Rich Hickey
Fleet Numerical

[Siekas, Greg]

Rich,

The "magic" is they replaced PXE boot with etherboot. If you flash with
the image from Dell you will only be able to PXE boot which will then
allow you to use something else besides Clusterworx and etherboot. So
yes, if you flash then you can't use Clusterworx, unless you happen to
have Clusterworx 4.0 / Advanced, then you can PXE boot the etherboot
image and use Clusterworx.

Greg

[Justin Wood]

I believe there's also a hack we used in the past to allow Clusterworx 3.x to boot using PXE.  I'd have to dig through my notes, but the premise is that in /tftpboot you create a directory structure that looks like what x-slam is trying to get, and in there you put a .zpxe file that'll do etherboot.  I've used it myself, but that was a couple years ago. It went something like this:

Example:
/tftpboot/
        x-slam:/
                192.168.18.6:10018/
Copy the [.zelf] binary into this directory _as_ the multicast channel.
Example:
        >cp   e1000.zpxe   239.192.0.128:10018

NOTE: The channel you use is going to be different on your machine and every
      time you create a new image and channel you'll have to do the same
      thing in /tftpboot for that channel.

-Justin.

[Eugene P Gooden]

With Clusterworx 3.4 you can PXE boot the etherboot image and use
Clusterworx.

It is a bit nasty in that you have to set up entries in:
/opt/cwx/etc/DHCPService.profile

We do it for our SuperMicro Visual nodes and whenever we replaced a Dell
or SuperMicro MotherBoard that LNXI had not flashed "correctly" first.

Still we are moving to using xcat (IBM) which is opensource now.

What we do with our IdataPlex nodes is to PXE boot a small image that
has GPFS included. With them booted we mount up the GPFS filesystem that
has a tar's "golden image" and a script that partitions the local disk,
creates the local filesystems, dumps the tar image, does a fixup of the
image based on node hardware type and function, does a mkinitrd, and does
a grub install. At this point it is a bootable disk image. The initial
PXE boot is from a xcat created image. More automation is about to be
intorduced in the near future.

regards,
gene

[Rich Hickey]

Thanks all. Looks like I'll need to do some quick research into
replacing Clusterworx. Bring these two systems into line with the
others we have. Either that or try one of the hacks. We're running an
older version of Clusterworx, so.....

Again. Thanks for the info, it's a great help.

Rich