[clug] SCP from Windows to Linux incredibly slow

[Paul Wayper]

At 08:36 PM 8/08/2004 +1000, Tony and Robyn Lewis wrote:
>On Sun, 2004-08-08 at 19:21, Peter Lavender wrote:
> > I have read the other replies, and most are looking at the network layer
> >   as a possible problem, which is a fair thing to do, however I suspect
> > that you will find it isn't the problem here.  (Of course I could be 
> wrong).
>
>Perhaps the following might be worth a try (though it's guesswork from
>me)
>
>1. does anything noticable happen to CPU usage on your win32 box when
>you use pscp?

No, the pscp or WinSCP process are using negligible amounts of CPU time on 
both machines.

>2. if you sniff packets between them, you might be able to see a pattern
>of who is being tardy with their responses

Good idea.  Let's see...

Well, interestingly, I got one transfer to go at 2MB/sec - all the rest 
seem to be sending one packet every five seconds or so.  I have no idea 
what the difference was, I'm just tracing through the packets in ethereal now.

>3. is there any unanswered DNS going on that it's timing out on?

Don't think so.  What's really odd is that it seems to start out going at a 
reasonable clip and then after two to three seconds slows down.  My best 
guess is that it's an emergent problem to do with Nagle's algorithm, which 
makes sense for character-by-character typing but somehow gets into a 
feedback loop when sending lots of data.

I read of an interesting attack on TCP where, by sending choke notices at 
exactly the right interval (around one packet per second) you could 
effectively DOS any port on a router by convincing it that the other end 
was saturated.  It's made me wonder if this sort of timing problem is involved.

>4. is putty's ssh similarly slow (you might test it by doing something
>like ls -laR / and seeing how fast it scrolls)?

Not at all.  Seems to be full speed.

>5. if you boot from knoppix, does the scp work better?

No idea - haven't download Knoppix to try it.

I'm getting Samba working between the two boxes, and that certainly moves 
data around like greased bowling balls on ice - i.e. much faster.  But it'd 
still be nice to know what the problem is - it should work...

Thanks all,

Paul

---------------------------------------------------------------------
SigClean 1.5a prevents all errors in transcription, transmission, and
reception of eMail, destroys sig and email viruses, and stops flames.
Report all bugs to the author. SigClean is FreeWare. (c)1998-2003 PJW