Getting a Linux diskless client running correctly

dave_v · 06-10-2003, 04:14 PM

I am trying to setup a system to be a diskless client. The machine in question boots via tftp (It actually comes up), but I have two problems (that I know now). The first is that the nfs mounts take an unusually long period of time to mount (~30 minutes) and the client doesn't know its hostname. Any suggestions?

zmedico · 06-10-2003, 06:24 PM

Have you tested the nfs from a normal linux system yet? Is it giving you any error messages in /var/log/messages? Are you using The Linux Terminal Server Project?

dave_v · 06-10-2003, 06:28 PM

We are nfs mounting two directory structures, and when I look at them (standard ls, cd, vi type commands) it seems to be okay. I'll check the log, is there anything in particular I should be looking for?

dave_v · 06-10-2003, 06:41 PM

I cannot remotely check the diskless clients because I can't logon, The server node (dhcpd host) shows a couple of things that may be related;

I see a DHCPREQUEST from the diskless nodes (I have 4 diskless clients I am trying to get going - this one appears to be x.x.x.220)
I see a DHCPACK (I assume a reply from my server - x.x.x.220)
I see a "Dynamic and statis leases present for x.x.x.217"
followed by "Remove host declaration hostname or remove x.x.x.217"
followed by "from the dynamic address pool for x.x.x.216/29"
followed by DHCPREQUEST/DHCPACK for x.x.x.217

Any ideas?

zmedico · 06-10-2003, 06:59 PM

Quote:

Originally posted by dave_v
the nfs mounts take an unusually long period of time to mount (~30 minutes)

Can you try mounting your nfs shares from a non-diskless client, just to see if it works in a reasonable amount of time from a normal client?

dave_v · 06-10-2003, 07:06 PM

I mounted the file system on the Linux server to my Solaris workstation and can see it (mounted faster than I could access it). The messages file on the Linux box shows an authenticated mount/umount request. I don't have a handy unrelated full up Linux machine to try a similar test with.

zmedico · 06-10-2003, 07:40 PM

Quote:

Originally posted by dave_v
The messages file on the Linux box shows an authenticated mount/umount request.

So the nfs on the server node seems to work fine. Are you using ltsp? Did you make the client kernel yourself?

dave_v · 06-11-2003, 08:25 AM

No I didn't make the kernel, It was supposed to be setup when we took delivery of these units. Just so I am straight, the kernel is an executable file that is the actual OS, and it exist in the / partion (right? - and what is the file's name?).

How do I find out if I am using Itsp?

What I have been doing to configure the downloaded file system is to modify a file system mounted in /loop and then make and a Makefile provided to cp, mv, mke2fs, e2fsck & gzip the new /tftpboot/rootfs.gz file system. This appears to work, but I am not sure about rebuilding the kernel (the last time I built a kernel was a few years back on a HP-UX system, and if my memory serves it was painful initially).

zmedico · 06-11-2003, 11:39 AM

Quote:

Originally posted by dave_v
It was supposed to be setup when we took delivery of these units.

Did you contact whoever sent you those units and if so could they tell you anything about them?

If your not using ltsp then you probably should be because they have already set everything up (including kernels) so all you have to do is install their packages. Just go to http://www.ltsp.org and get the packages for whatever linux distro you want to run on your server node.

There is a great explanation of how the diskless clients boot here:
http://www.ltsp.org/documentation/lt...-en.html#AEN53