[PLUG] Challenging Problem . . .

[drew wymore]

On Mon, Oct 11, 2010 at 10:06 PM, frankhunt <fh-linux at comcast.net> wrote:
>  Situation: Client = Vista PC, NTFS file system; Server = CentOS 5.3
> serving CIFS shares (FYI: neither of these systems are current with all
> the latest patches and updates - customer preference).  Client has
> windows open on both systems and attempts to drag and drop files from
> server to local file system.  Seems to work on smaller files.  Files
> larger than about 3-5MB will stop transferring after about 10% of the
> file and hang.  Client times out, complains that it can't see the file
> on the server - the network connection to the server is lost.  Cannot
> ping server from client or client from server.  Cannot ping the router.
>   Run "ifconfig eth0 down"; "ifconfig eth0 up" on the server side and
> the network comes back.  Transfer failure is repeatable (usually)
> although sometimes it works.  Network is straightforward SOHO type -
> D-Link wireless router, Netgear switches, very light traffic.  I do not
> have a topology yet.
>
> Approach:  Added my laptop (HP Pavilion dv6/Ubuntu 10.10) as a client
> and the problem persisted.  Same symptoms transferring files to/from the
> CentOS system.  Transfer would start, hang, time out, the network would
> crash and re-appear after bouncing the network on the server.
>
> Connected the laptop directly to the server with only a Netgear switch
> (no router).  File transfers would NOT fail.  Could not reproduce the
> problem in this configuration. [aha, he said, it's in the network!]
>
> (but wait)
>
> Replaced server with the laptop, served files to PC via CIFS.  We would
> expect the laptop to fail just as the desktop CentOS system did (because
> we think it is a network problem).  Wrong - works perfectly.  Could not
> reproduce a failure in this configuration.  [Aha, he said, it's NOT the
> network]
>
> Now what?
>
> When I left my client this afternoon at 5PM, we were having success by
> using "copy and paste" instead of "drag and drop" on the PC.  WHAT?
> Why does this work?  We will continue to test this transfer process but
> I don't see why it would make any difference (but hey, it's Windows on
> the client).  Have not tried this approach using the Linux laptop as client.
>
> Thoughts:  Could be some buggy condition in the server software
> somewhere (I suspect this because the system has not been updated but
> have no real reason).  Could be some issue with the CentOS drivers
> interacting with the network components.  I'm really saying that I have
> no clue.
>
> Hence this post.
> Looking for ideas, approaches, incantations, spells, band-aids, anything.
>
> Thanks for listening,
>
> --
> frank hunt
> (L0F) R0B-ZAR1
> befuddled linux admin
> erstwhile photographer
> hillsboro oregon
>
> _______________________________________________
> PLUG mailing list
> PLUG at lists.pdxlinux.org
> http://lists.pdxlinux.org/mailman/listinfo/plug
>

Frank -

Look at the NIC on the Cent box and see if it starts incrementing
errors, check dmesg etc to see if there is anything obvious logged.
The copy and paste vs drag and drop is weird if the D-Link is in
place. I ran into a similar problem between Windows clients with a
wireless AP in place where one was connected to the AP via G and the
other via N and running wireshark (works on windows and linux woohoo)
and saw the buffers filling up causing the stall. This might be a
worthwhile route to go to ferret out the problem and see if you can
see a difference with the AP in place between drag and drop vs copy
and paste.

If the AP lets you, maybe check the WLAN interface for errors as well.

Drew-