[PLUG] Incredibly Slow scp Across Local Ethernet

[Kyle Hayes]

On Monday 12 January 2004 17:20, Paul Mullen wrote:
> On Fri, Jan 09, 2004 at 02:21:42PM -0800, Kyle Hayes wrote:
> > I know several people who've seen this.  If the driver is a module, try
> > rmmod-ing the driver and reinserting it.  If that causes data to flow for
> > a
>
> This didn't seem to work for me. Am I supposed to be able to rmmod the
> driver while it's being used? rmmod complained the module was busy.

You'll have to close all connections and bring the network down at least 
before the module is freed up.  After that you need to make sure that no 
servers are tying things up (Apache on the NIC IP or something).

> > The other case I found was a NIC that would only do 50kB/s upload. 
> > Downloads were just fine (100mbps).  The odd asymmetry was something I'd
> > never seen
>
> ---8<-----
>
> > Can you determine if the problem is in one direction or both?
>
> That's really weird. I tried a suitably large file in the other
> direction (PPC to P120), and the transfer flew right by. Great
> throughput.

Hmm, so it does seem to be dependent on direction?  Is that what I'm 
understanding?

When I was trying to figure out my NIC problem, I tried swapping cables etc. 
Nothing made a difference but the NIC itself.  Outbound, 50kB/s, inbound lots 
more.  See if it isn't a flaky cable.  If you can just reverse the cable and 
get the reversed results, you might have a flaky cable.  That's cheaper than 
a NIC :-)

> > What are the drivers and the cards?
>
> The P120 is using an old Megahertz 10BT/modem combo-card. It uses the
> smc91c92_cs and serial_cs drivers. AFAIK, the modem is treated as an
> entirely separate device. Both modules are loaded at the same time,
> though. The PPC has a 10/100/1000 MB ethernet adapter, which
> uses the "sungem" module.

Do you have any other machines you can test against the laptop and PPC 
machine?  It could be the PPC machine that has a problem with inbound 
traffic.

SMC makes one of the chipsets that apparently has the problem with high 
bandwidth according to someone I know.  Which chipset it is, I don't know, 
but it was fairly old (six or seven years I think).  Can you check the 
interrupts and make sure that the ether isn't sharing with anything else?  
Laptops often have some weird interrupt assignments.

Note that the asymmetry I saw was not latency, but bandwidth.  Latency was the 
same in both directions.  It was weird.  Maybe the card was going bad?

Best,
Kyle