[SLL] How to troubleshoot corrupted wget/apt-get/bittorrent file transfers?

[Phil Mocek]

On Mon, Mar 09, 2009 at 06:26:07PM -0700, Kurt Buff wrote:
> On Mon, Mar 9, 2009 at 18:20, Phil Mocek <pmocek-sll at mocek.org> wrote:
> > It looks roughly like this: Cisco 678, with an old Webramp something
> > running Sonicwall SOHO firmware, both in a nest of power cables,
> > then CAT5 cable from there running over doors, across floors, and
> > down a handrail, around the corner, over some doors and through a
> > couple closets to a desk where there's some no-name switch buried in
> > a pile of network cables, power cables, and various other gear.  Not
> > pretty.
> >
> > But these are TCP connections.  Isn't any junk on the wire supposed
> > to be cleaned up at the transport layer?
> 
> Heh. Not necessarily, though in theory yes. If there's an issue in the
> network, though, you'll likely see lots of retransmits and other
> errors.

I think there are multiple issues at play, at least one of which is a
problem.

I fired up Wireshark, told it to watch eth0, and saw tons and tons
of errors scrolling by.  I stopped after a few minutes.  "Analyze
-> expert info composite" reports two errors: about 25,000 TCP bad
checksums and about 1400 UDP bad checksums.  It also reports
warnings including unreassembled packets, previous segments lost,
out of order segments, ACKed lost segments and fast
retransmissions.  I'm no network engineer, so to me, this all just
looks like "bad".

So, what now?  Run wireshark on a laptop, moving one step at a time
along the wired net from the ADSL modem to the switch at the end until I
see a big increase in trouble, then do some wire maintenance?

-- 
Phil Mocek