[PLUG] Slowest Full duplex 100MB in the west

[Don Buchholz]

Jeme A Brelin wrote:

> Overnight, I left the FTP running and the errors tripled on the 
> interface.
>
> So I just restarted robot and included the "tulip full_duplex=1" in 
> /etc/modules (though there was no tulip line there even though the 
> module is listed in lsmod).
>
> Everything seemed fine.  Counters reset.  No errors.
>
> Fired up FTP... no errors.  Started a file transfer.  Woah!  Several 
> errors a second.  Killed ftp client.
>
> Fired up a flick on the XBox.  No errors.  Nothing but smooth transfers.
>
>> You'll also note that he has frame errors.  Frame errors are when the 
>> received packet has a bad CRC and/or non-integer number of octets. 
>> These are usually the result of collisions or bad Ethernet device.
>
>
> I really WANT to blame my crappy switch (so I can justify buying a new 
> one), but I'm not seeing ANY errors on my new desktop machine or 
> transfers between other hosts.
>
> Let me run an ftp TO the XBox from person.
>
> Dig this.  400KBps.  No way to check interface errors on the XBox, but 
> my guess is that we will find few to none with a transfer rate like 
> that. (Still slow, though... the switch lights green or orange 
> depending on whether the interface is 10 or 100Mb and the XBox is 
> orange.  I replaced my last internal green-lit card last week.)
>
>> Jeme's situation, and that of most home users, is a bit trickier 
>> because he/they don't have managed switches.  The typical home 
>> Ethernet switch is auto-sensing and the user has no way to set port 
>> parameters or view port statistics.
>
>
> And managed switches are really expensive.
>
>>> I'd guess crappy cables, or *serious* interference, or
>>> something similar on the physical layer...
>>
>>
>> That could well be, that's for the person on site to evaluate.
>
>
> I'm more inclined to think it's the switch or the NIC than the cables, 
> but only because I would think the cables would munge data regardless 
> of transfer speed.
>
A couple of years ago, I had an ethernet switch/hub (built into a IDSL 
router) fail
in a peculiar manner.  Small frames worked just fine.  I could 
ping/traceroute all
day long without error ... until I started to increment the packet 
size.  POP3 users
accessing small messages didn't even notice a problem ... those who had mail
messages over a few K (esp. those w/ multi-meg PDF, etc.), or were trying to
use an FTP server on the same segment were completely shutdown.

First, make sure you have a quick test which reliably reproduces the 
problem.
(And it sounds like your FTP test is working well.)

Then, the quickest (and cheapest), thing to try is to use a different 
port on the
switch.  (For that matter, what happens if the "robot" and "person" 
ports are
swapped?)  You may just have a bad port, and I've seen more than a few
engineering/lab nets with red tape covering blown ports....

Second, swap out the cable ...  again, fast, cheap.

Next, you get to decide if you want to swap out (a) the switch, or (b) 
the NIC.
....