Display Problem with PelicanHPC v1.8 64bit

[seanbegley]

Hi all,
I'm currently trying to get PelicanHPC working on an Intel Xeon based machine. I'm using the standard iso from the downloads page but am running into a problem duing booting. Every time the xserver tries to start, the PC reboots. I have an ATi FireGL 7350 graphics card.

I have tried using different cheatcodes to change the xmodule being used by the xserver, but all seem to cause the same problem. I think the problem is with X because I'm able to get to a command interface, and it's only when i try to start X that the PC reboots.

I have also tried a Debian 5.0.0 live CD to see if I could figure out what to do. The Debian 5.0.0 boot up process runs to the point where X is starting, and then the screen goes blank, no matter what xmodule I tell it to use. It doesn't cause the PC to reboot though.

Any ideas on what could be causing these problems, is the graphics card compatible with the standard ati driver?

Thanks,
Seán

[Michael Creel]

I don't know the cause or solution, but the last release of PelicanHPC is essentially a Debian Lenny live CD, because it's made using a Lenny base just before release of Lenny. I suggest searching for problems with your video card associated with Debian Lenny, and possibly asking on the Debian Live mail list.

I'm starting to think that the previous behavior of booting to the console, with "startx" used to go graphical, might be better than the current behavior. At least that way it's pretty much guaranteed the cluster can be set up.

[seanbegley]

I'll keep working at it and will post on the Debian forums. I'll keep you posted on this forum as I make progress (if any!)

Thanks again,
Seán

[Michael Creel]

Seán - there is an iso image in the normal download area marked as v1.9 beta1. This image boots up to the console, and then X is started by typing "startx" (you probably will not be able to use X on your machines). From the console, just log in (username "user", password whatever you set). Then set up the cluster by typing "pelican_setup". I'm quite sure that this will at least allow you to set up a cluster. If you try it, please comment here as to whether or not is works.

[seanbegley]

I will try it out now. and let you know this afternoon how it went.

Thanks

[seanbegley]

Unfortunately this also didn't work. Logged in at the prompt without any problems and ran the pelican_setup script, all grand. When I PXE booted the compute node, it stopped booting at the network card detection. It got as far as:

[ 14.xxxxx ] tg3: eth0: Link is up at 1000Mbps, full duplex
[ 14.xxxxx ] tg3: eth0: Flow control is on for TX and on for RX

at this point it freezes.

I know that using the older PK distribution we had to select the network card driver from the list of drivers during initial cluster set up, the current freezing may be related to that.

If I find a workaround for any of these problems i'll post them up on the forum.

Seán

[Michael Creel]

What a pain! I guess that if I were in your situation, I would go out and buy a few Debian friendly NICs, which would probably be the fastest and cheapest solution. I have to confess that my plans for PelicanHPC do not involve using anything but stock Debian kernels, so NICs that require nonfree modules are liable to remain a problem.

[Michael Creel ()]

One last idea: does the machine that needs to PXE boot have more than 1 ethernet port?. If so, try switching the cable to the other port. I believe that compute nodes like to receive things on eth0, not eth1. See the thread http://www.nabble.com/Node-boot-problem-td21355444.html , the last few messages.

[seanbegley]

Phew, I finally got my graphics card to work. I messed around with the xorg.conf file for the last few hours until I figured out what to do. If you have an ati graphics card there are a number of drivers that ati can use.

In the Device section of the xorg.conf file, if you specify the driver as "ati", the correct driver is meant to be chosen. The "ati" driver is really a wrapper for a collection of drivers, and the wrapper is meant to choose the correct one for you.... not the case.

I ended up trying the other relevent options {radeon / radeonhd / r128 / atimisc} along with a rake of other options including {vesa / vga / chips......}

In the end, it turned out that the correct driver for the Ati FireGL 7350 is radeonhd.

There's no need to change the xorg.conf file by hand, you can specify the driver when booting using a cheatcode. Of course, you may want to change the display modes available, in which case you'll have to change the file.


boot: live xdriver=radeonhd


One problem down.... a few to go!

Seán

[seanbegley]

Even though I have the compute node up and running with and sfce window now, the compute nodes still won't boot. The compute nodes only have one ethernet port, and are still hanging at the same point. Is there a cheatcode that I can input at boot time to disable the ipconfig script? I have tried "nonetwork" and "nodhcp" but with no luck. Or even better, is there a way that i can copy the correct ipconfig file over to the compute nodes?

Seán

[Michael Creel]

There is a way to pass IP, gateway, etc. at boot time (consult the live-initramfs documentation) but that should not help, because to get the cluster to work you would just be passing the same information that the server provides. Sorry, but I'm out of ideas here - I was hoping the multiple NIC thing was the problem.

[Michael Creel]

I have found some information about how to add the NetXtreme firmware to the initrd that is used by the compute nodes:
http://wiki.debian.org/DebianLive/FAQ#head-0e5757bf06d76e8feb9e78ffa2d1b3b442868224

Incorporating this into the make_pelican script looks like a non-trivial amount of work to me. I think it can be done, but I don't have time to try it now. The solution of buying different network cards seems more attractive than trying to do this.

Good luck, M.

[seanbegley]

I figured out that the compute node wasn't really hanging when the network card was being set up. The timeout for the card is high, I don't know the value yet. A long story cut short is that I left the net-booting PC alone, went off for lunch, and when I came back it had fully booted! So now I have a masternode with a working graphics card and a net-booted compute node!

Thanks for all of the help.

All that is left to do now is to install the new software and re-master the cd.

All of the newly added software on the remastered cd will be available on the compute nodes, won't it?

Seán

[Michael Creel]

Great! Note that you don't actually remaster the CD image as you did with PK, instead you use the live-helper package and the make_pelican script to make a new image - see the Tutorial for more explanation.

The other possibility, if your software is not in Debian, is to just install the new stuff into /home/user, and the compute nodes will have access. To make this reusable across reboots, you can put /home on a hard disk partition (ext2 or ext3). Again, see the Tutorial.

Cheers, M.

[seanbegley]

Hi Michael,
I've tried out v1.9beta2. I'm still having the same problem with the compute nodes taking a long time to boot because of that ethernet card issue. The compute nodes all work once booted, it just takes about an hour for them to get past the ethernet card check.

Other than that, I no longer have any issues with increasing the size of the ramdisk beyond 512MB, (I currently have it at 2GB), adn no longer have any display issues.

I'm now at the stage where I can use the cluster for some of the tasks that I have to do, so it's all going fairly well. The final thing that I have to do now is to make a new cd with just two additional packages.... bc and libxp6. The first attempt at using the make_pelican script failed, but because I just left it running at the end of the day and came back to it in the morning, I have no idea why or where it is having problems.

I will try and add these packages again to the make_pelican script and see how it goes. Should I have the make_pelican script in the root directory ( / ) ?

Seán

[Michael Creel]

I don't know what's up with the long time the nodes need to boot, I've never experienced that. Maybe a general search for tips with PXE for that card might turn up something.

For building a new ISO, you need to install live-helper and debootstrap, and you need to run make_pelican as root, from a directory that has a LOT of free space available (several GB, to be on the safe side). To have MPITB included, you would need to copy the stuff in the Econometrics directory of a running PelicanHPC into a pelicanhome subdirectory of the directory in which you run make_pelican.

That's a little complicated. I'd be happy to make the image you need, if you prefer - it's virtually no work for me. You cited one of my papers, so you get special treatment

[seanbegley]

I'd be delighted if you could make the new image for me, the second attempt at using make_pelican failed.

It failed to retrieve the release. I guess I have to check my repository sources list and make sure that apt can connect to each repository. I'll figure out what's gone wrong with this as an academic exercise. It wouldn't hurt to know.

Thanks again for all of your help, it is really very much appreciated.

Seán

[Michael Creel]

Hi Seán,
There is a v1.9 beta3 that has those 2 packages in the download area.
Cheers, Michael