show stopper for us: dev nbd0 clients locking up losing connection with server

[john]

Hi Scott,
>
> Well, if you're not doing a huge amount of stuff with graphics in firefox or
> Gimp, you probably won't need the swap, so I'd turn it off if it's making
> things worse.

Ok, thanks. Is there a place I can go to get a feel for how ltsp
currently works? Is there a doc/web page etc that
sort of says here's how we implement LTSP since 7.10 and here's some
options you can work with e.g
you can edit  "/etc/ltsp/nbdswap.conf" etc?

Sorry I am having a hard time getting a handle on the move to NBD etc.

> Here's the specs of a machine that's running as I type this:
>
> 2 Dual core Xeons 5140  @ 2.33GHz
> 8 gigs of ram
> 2 intel ethernet cards.  It's an IBM x3550 server box.

Seems like we have pretty similar installations...

>
>  All of them are using NBD.  I'm using Via E10000 mobos with 1 gig of ram
>
>  So, I know the NBD works, at least in my configuration.  Now, I've got "dumb"
>  switches: i.e. non-managed regualar D-Link switches.  I wonder if something's
>  going on with the networking side of things.  Do you have Managed switches,
>  and if so, can you get some stats out of them?


Sadly most of them are dumb as well. I believe one may be able to do
logging remotely. I'll look into turning that on.

One intersting note:

When I showed a *nix guru I know the log messages :


Sep 12 10:34:44 192.168.0.248 kernel: [679871.912346] end_request: I/O
error, dev nbd0, sector 200140
 Sep 12 10:34:44 192.168.0.248 kernel: [679871.912363] end_request: I/O
 error, dev nbd0, sector 200142

He pointed out "those sector numbers are just a hair above 50 megs.  Have you
looked at the block files which are being served for the clients (I believe
they're in /tmp)?  I wonder if the files aren't being sized correctly for
the swap space they need to serve."

Dunno if this rings a bell with you, but I"ll be look into that next.

Thanks for any ideas you may have!

John