TR Forums

Do you have any links / suggestions for getting LTSP setup to netboot nodes? I've got about 35ghz worth of Dell Optiplexes sitting in a corner (unused) That I would to get folding, if I can make them all netboot.

Thanks,
Will

retro2001 wrote:

Do you have any links / suggestions for getting LTSP setup to netboot nodes? I've got about 35ghz worth of Dell Optiplexes sitting in a corner (unused) That I would to get folding, if I can make them all netboot.

Thanks,
Will

If those Dells have CD-ROM drives then you can use notfred's Folding CD generator.

Sadly, these Dells have no CD or Floppy drives, and it would take too much time and effort to clone out their hard drives, so pure netboot it must be.

retro2001 wrote:

Do you have any links / suggestions for getting LTSP setup to netboot nodes? I've got about 35ghz worth of Dell Optiplexes sitting in a corner (unused) That I would to get folding, if I can make them all netboot.

The instructions at the main LTSP site are pretty good as far as they go. They are more geared towards using the netbooted systems as X-Terminals though. I can probably walk you through the process of getting everything set up for diskless folding. (Actually it is more correctly called "shared disk" folding, since what you're effectively doing is having all of the netbooted nodes share a single hard drive on the LTSP server.)

Do these systems have integrated (on the mobo) NICs, with a netboot option in the BIOS? If they do, it should be relatively straightforward to get things set up.

Failing that, if they have floppy drives and you can figure out what chipset the NICs are using, you should be able to create a floppy-based Etherboot loader using ROM-O-Matic.

If they have CD/DVD drives, Flying Fox's suggestion is probably the way to go instead.

If you're up for giving LTSP a go, the first thing you'll need is a server running some flavor of UNIX. It doesn't have to be anything fancy; even an older, slower machine will be adequate since F@h is almost entirely CPU-bound, and doesn't stress the server. (My "Mark I" crate farm had a 1GHz Athlon as the server, and I suspect I could've netbooted a couple dozen nodes from it if I'd wanted to.) I'll probably be able to provide the most assistance if you use Fedora, since that's what I used myself.

You also need to be able to serve DHCP to the diskless nodes from your LTSP server. This means that if there is another existing DHCP server on your network, you need to either prevent it from responding to DHCP requests from the diskless nodes, or put the diskless nodes on their own subnet (behind another router) to prevent the DHCP requests from reaching the existing DHCP server. On my home network, I deal with this issue by using static IPs for everything except the diskless nodes (DHCP is turned off on my router); this may or may not be an option for you depending on whether you need to integrate into an existing LAN, and how your LAN is set up.

Post some additional details about the hardware and your network, and we can take it from there.

Edit: OK, no CD or floppy. What about the integrated NIC? Failing that, can these systems be configured to boot from a USB flash stick?

Flying Fox wrote:

If those Dells have CD-ROM drives then you can use notfred's Folding CD generator.

Just see my site in my sig for netbooting with a Windows machine as a server.

retro2001 wrote:

Do you have any links / suggestions for getting LTSP setup to netboot nodes? I've got about 35ghz worth of Dell Optiplexes sitting in a corner (unused) That I would to get folding, if I can make them all netboot.

Thanks,
Will

NotFred has a great "How To" guide for diskless folding with Windows XP.

You can also use the NotFred Diskless Folding CD and use one (1) CD-ROM to start up each PC. Just disconnect the power plug & IDE cable for the CD-ROM after your diskless PC is folding. I've done it without any problems.

You can also try the OverClockix LTSP 2.1 to set up a server and diskless folding farm. The LTSP 2.1 is already configured for diskless folding. You can ask me or Cass for tech support.

And while I'm here, I'll ask you to join UnitedGerbilNation. 35 Ghz of folding power would catapult the Nation into the #2 position on Team TR. That would be fun.

They are Optiplex 170L's with integrated Intel 10/100 Pro NICs that claim to support netbooting. The BIOS options and bootup process surrounding the netboot ability is odd. I've tried (and failed) to get them to boot into PXES, and watching my DHCP server's log, it didn't look like they even requested a DHCP lease (and such) before giving up on the netboot. Does Dell have some kind of special requirements for netbooting systems like these? Also, at my disposal, I've got a horde of Windows boxes and 2 linux servers, so my options as far as netboot servers go are pretty wide.

An aside, my recent jump up to 2500 PPD (a 500 PPD increase) is without any of these boxes, and I think that my total should (again sans these boxes) stabalize up around 2700....The race to 50 is not going to be won easily....

I don't know one way or the other regarding special requirements to netboot Dells. But if they support PXE, it should work. Can you give any more details on what the netboot BIOS settings look like?

Maybe the netboot firmware on the NIC needs upgrading... might be worth checking if there's anything available for download.

Seems odd that they don't request a DHCP lease. Is it possible that the DHCP server is configured to only respond to specific predetermined MAC addresses?

Can the systems be configured to boot from USB media?

They can be setup to boot from USB media, I believe. There are no real options for netboot in the bios, aside from an option for allowing WOL being "Wake then boot from network". To boot from the network without waking the computer from the LAN you have to press F3 on startup, at which point the computer says (in the top right corner) Attempting Network Boot... and then after a moment, continues on booting to other media (or failing if there are no other options). There is no error message or even status message during the Net Boot attempt. The NICs are onboard, so I don't know if their firmware can be updated, but I have installed the most recent BIOS, without gaining anything from it. My DHCP server is set to allow any computer to request an address, and actually even had an entry specificly for this computer with the boot parameters.

I've been unable to dig up much in the way of details surrounding exactly what protocals these support netboot-wise. Dell's documentation is quite sparse.

cheesyking had a post that belonged both in this thread, and the one it was split from. Here's the relevant bit:

cheesyking wrote:

Retro, any idea what version of PXE those Intel NICs have? I found I had to use V2 or later to get it to work (though they were still picking up an IP so it's probably a different problem). There is a section on etherboot in that guide I linked to, could be worth a look.

I suck at the BIOS (somewhat). The BIOS update added a new option to the network adapater's setting (on / off / on with PXE <-- new) that I hadn't noticed. It's PXE Version 2.1 with Intel Boot Agent 4.1.06.

With this enabled, the netboot keypress brings up the more familar looking Intel Boot Agent... and the computer requests DHCP info and whatnot.

Whoo...I'm on fire today.

If you are happy with Linux, then just grab the relevant files from my Windows diskless folding guide and put a TFTP server on Linux. It's a custom build which boots and folds and that is it. There's a tiny webserver to see the logfiles as well.

Is your DHCP server currently one of the Linux boxes? (It'll make things easier if it is.)

Next step is to install LTSP 4.1.1 on the DHCP server and get the diskless node to boot. By default it'll boot into an X-Terminal session, which is not what we want... but that's relatively easy to fix once we've got the basic LTSP infrastructure working.

I'm still trying to work out the best way to get F@h launched on the diskless nodes. I've gotten it to work by running commands on the diskless node remotely with ssh, but that's a can of woms because you need to set up both NIS and SSH (more stuff to go wrong). I suspect that there may be a simpler way...

All my notes on this stuff are at home right now, so unfortunately I can't go into detail on all of the issues I encountered until I get home this evening.

Edit: Or in the meantime you could try notfred's suggestion...

Last time I did LTSP folding I mounted the folding directory for that client and just tacked something like


onto the end of LTSP's rc.init script

Real quick, but real dirty

cheesyking wrote:

Last time I did LTSP folding I mounted the folding directory for that client and just tacked something like

Code: Select all

cd /fah
./FAH502-Linux.exe -forceasm

onto the end of LTSP's rc.init script

Real quick, but real dirty

Yeah, I was thinking of trying something along those lines instead of the SSH/NIS approach. But you lose the ability to monitor the diskless system or do a clean shutdown of the folding client in the event that you want to power the node off.

Edit: Also, how would you handle multiple nodes? Can each node have its own rc.init?

Edit #2: I suppose you could just name the directories based on the node name and use the node name to cd to a different directory for each node.

just brew it! wrote:

Yeah, I was thinking of trying something along those lines instead of the SSH/NIS approach. But you lose the ability to monitor the diskless system or do a clean shutdown of the folding client in the event that you want to power the node off.

Edit: Also, how would you handle multiple nodes? Can each node have its own rc.init?

Edit #2: I suppose you could just name the directories based on the node name and use the node name to cd to a different directory for each node.

Loosing monitoring is a bit of a pig, though doing that does mean the client's output goes to the display on the node so you can plug a monitor in to help debugging.

multiple nodes are dealt with by mounting each clients' own folding directory on fah, does mean a little extra work on the exports file but not that much.

cheesyking wrote:

Loosing monitoring is a bit of a pig, though doing that does mean the client's output goes to the display on the node so you can plug a monitor in to help debugging.

I plan to stuff the crates into the crawlspace once they're working, so being able to use the console port for debugging doesn't really matter to me.

multiple nodes are dealt with by mounting each clients' own folding directory on fah, does mean a little extra work on the exports file but not that much.

How would you do that? I thought it was the NFS client that determines what gets mounted where, not the server?

I've got one of the beasts up and folding courtesy of notfred's magic boot image. Provided it holds together, I hope to get the party started this weekend...

retro2001 wrote:

I've got one of the beasts up and folding courtesy of notfred's magic boot image. Provided it holds together, I hope to get the party started this weekend...

Cool.

Looks like this is going to be a good week for TR adding firepower!

I just got the first crate of the MkII Crate Farm rebuilt as well. I'll post some pics back on the other thread (the one which spawned this one).

Glad to hear you got it working. Did you remember to configure the TFTP server to let it write the backups back to the server?
BTW it should hold up on the folding, I've basically been running that image since last Christmas on my diskless folder. If there are problems, I would check the hardware first.

retro2001 wrote:

I've got one of the beasts up and folding courtesy of notfred's magic boot image. Provided it holds together, I hope to get the party started this weekend...

Good work ?Retro!!! Nice to see issues being resolved.

just brew it! wrote:

cheesyking wrote:

Loosing monitoring is a bit of a pig, though doing that does mean the client's output goes to the display on the node so you can plug a monitor in to help debugging.

I plan to stuff the crates into the crawlspace once they're working, so being able to use the console port for debugging doesn't really matter to me.

multiple nodes are dealt with by mounting each clients' own folding directory on fah, does mean a little extra work on the exports file but not that much.

How would you do that? I thought it was the NFS client that determines what gets mounted where, not the server?

You use variables and the ip of the client to determine the mount directory determined by the name served from the pxe/dhcpd.

http://www.extremeoverclocking.com/articles/howto/FAH_Diskless_Farm_4.html

The above link gives the information... It is old, so I don't know how relevant it is now.

yeah my exports file looks something like:



Because I'm not using LTSP I have to create a pxelinux.cfg file for each client to get it to boot from the correct root directory, but using LTSP gets round that bit.

EDIT: forgot this bit

You'd then need to setup fstab in each client's root to mount the correct dirs. Using LTSP you could forget about fstab and just add a mount command to the rc.init file (using the workstation name variable (in my case node1,2,3) to select the correct working directory)