PXE boot and kernel module

[CanAm]

All,

The master and compute nodes have an ethernet card which is not supported by the kernel in Pelican HPC (e1000e). For the master node I downloaded the driver, make, make-install, modprobe etc. and managed to get it up and running.

Do I need to do anything to the PXE side of things so that the compute nodes (as they also use the same network card as the master node) also receive this driver? If so how is this achieved?

Thanks in advance.

[Michael Creel]

Which version of Pelican are you using? For 1.8, I attempted to include drivers for nonfree NICs. I don't know the details about the e1000e driver - what are the relevant Debian packages? I thought Pelican already worked with all  e1000 cards, but I guess there's a version I don't know about. In principle, adding the package names to the package lists in the make_pelican script and then making a new image should solve the problem.

[CanAm]

I tried it with the v1.8 32 bit.

On the master node I installed the e1000e driver and then ran pelican_setup from a console; I could not get X windows to work as my display gave an error messaging indicating it was out or range.

The debian packages required to build the driver are linux-kernel-headers-2.6.26 and for good measure I downloaded all of the other dependencies also (i.e. cpp, gcc, libc6, libc60dev and so on). The source code for the driver was downloaded from http://sourceforge.net/projects/e1000/. Once the packages to compile the kernel were installed, it was trivial to make, make_install, modprobe e1000e as root.

When at the appropriate stage, I began to switch on the compute nodes to allow them to boot via PXE.

However the displays attached to the compute nodes began to show the splash/boot screen identical to that which was shown on screen attached to the master node when it was booting from the live CD (i.e. a black/white image of pelicans). It then asked how I wanted to proceed with the boot sequence (i.e. Press F1 for help or Enter to boot:). This behaviour appears to be quite different to that shown in the tutorial web page.

Finally, during the boot phase of the master node, the console gave many I/O errors for a sector on the CD. I verified the MD5 of the downloaded image and then burnt a second CD at a slower speed. However this new CD also gave similar I/O sector errors. As an alternative solution, is it possible to boot pelican from a USB stick, and if so what are the steps involved?

[Michael Creel]

Are you sure that plain Pelican 1.8 does not work? I don't know the details about this NIC, but Intel NICs usually work without any trouble. The I/O errors you see suggest that your CD/DVD drive may be dirty/loose cable/etc. It is possible to use a USB stick, just search this forum for "USB" and you'll get some hits. However, using USB will still leave you with the NIC problem, assuming that plain Pelican will not work.

I think there will be trouble trying to get the nodes to PXE boot if the driver must be compiled from source, since the kernel supplied to the compute node will not have the driver, and it will be difficult to figure out how to provide it to them. This situation does not fit into the scheme used by Pelican, and I really don't have a good idea about how to work it out.

Seeing the boot splash on the compute nodes is normal, you don't need to touch them, they will boot up after a brief period. The reason there is a pause is that it gives the opportunity to pass boot parameters.

Good luck!