Network Card problem

Gavin McCullagh gmccullagh at gmail.com
Wed Oct 25 09:42:51 UTC 2006


Hi,

On Tue, 24 Oct 2006, James Call wrote:

> First of all thanks for your responses about clustering.  I can see
> the point that it is too much hassle for the small benefit.

To be fair, I'm working off instinct not real benchmarks.  It might work
well, I only suspect it's too much hassle for the benefit.

> Second.  Does anyone have fat clients working?  I guess that I don't
> understand the terminology.  What is the difference between a diskless
> client and a fat client?  

My understand is there are three likely options:

1. Traditional desktop with OS installed to disk.

2. Diskless thin client. No disk, minimal OS comes over the network,
   sufficient to start an X server and offer a display.  Thin client server
   runs programs which display on the thin client.

3. Diskless workstation(?).  No disk, complete OS and installed programs
   come over the network as required (usually over nfs).  The workstation
   itself runs all of the programs using its own cpu & ram.  However all
   disk reads and writes are to network drives.

[1,2] should work trivially.  I have not done [3] myself yet so I can't
guarantee but I'm told it works well.  It has definite advantages over [2]
in that the programs can directly access sound card & usb keys, there's
less cpu usage on the server, (mostly) reduced network load, etc.

> We are using p3 clients without hard drives.  They have a ton of CPU
> capacity that goes unused, while the server is getting dogged down.

Sounds like a good candidate for [3] alright.

> We have 2 nics one with address 192.168.1.100 that serves DHCP to the
> clients in the lab.  The other nic is 192.168.2.151 that brings
> internet in from the office.  When the clients boot, they hang like
> they are getting another DHCP offer from a router or something.  

You should confirm this by putting a machine on the network and running
something like:

	tcpdump -i eth0  | grep BOOTP

which should show you all of the dhcp responses.  Just to make sure you're
chasing the correct problem.  Alternatively set the password on the thin
client, login to it at the text console and run "ifconfig" to see what
address it got.

> So I turn of nic2 (192.168.2.151) and the clients work.  

You obviously have two subnets: 192.168.1.0/24 and 192.168.2.0/24.  Are
they on the same physical wires?  If so, you must be very careful running
two dhcp servers (or not do it at all).

1.  The client sends the DHCP request as a broadcast (to all subnets).
    After all it doesn't know what its subnet is.
2a. The dhcp server on 192.168.1.0 answers with a 192.168.1.0 address.
2b. The dhcp server on 192.168.2.0 answers with a 192.168.2.0 address.
3.  The client gets two responses and picks one or other, which is not well
    defined.

> Then I boot everybody up.  Then turn on NIC2 again for the internet.
> This is a totally stupid workaround, but I couldn't figure anything else
> out.  I can't find any other box offering dhcp.  

If this is fixed by unplugging your server's second interface, this
suggests that either your server _is_ the second DHCP server, or that it is
forwarding DHCP requests (which would be very strange).

> If I disconnect the cable from NIC2 reboot the server, the clients hang
> at the same point.  So it isn't a DHCP conflict.  

I don't think I follow your logic here.  Have you not just found another
way to hang them?

> I am guessing it is something crazy within the server.  I spent about 20
> hours trying to figure out the problem.  

If you think it's not a DHCP problem you need to give more information.
What sort of hang?  Do they boot at all and if not what point do they stop
at?  Do the already booted machines hang?  In what way?  Does the server
continue to work, is DHCP still working?  If you boot a machine what
address does it get?

> Then in frustration I reinstalled (and updated) to 6.06  That was a drag
> to add all the users again?!@#$#@#@

If you need to reinstall, you should be able to back up /etc/passwd,
/etc/shadow, /etc/group and take a tar of /home.  When you reinstall you
should then be able to restore all users by replacing those.

> I still have this problem.  The worst part is that I am not the
> computer teacher, so I had to make a training manual showing how to
> use webmin to take down then up the NIC2.  There has to be some
> reasonable solution to this.  Any ideas?

The reinstall didn't fix this then.

[ I would have thought it would be easier to just teach them the ifconfig
command to be honest, but whatever you prefer. ]

Gavin





More information about the edubuntu-devel mailing list