Re: scp/ssh between two cygwin installations very slow

[Bernhard Ege <bernhard at ege dot cc>]

Larry Hall wrote:
> At 06:58 PM 4/25/2005, you wrote:
>
> > Christopher Faylor wrote:
> >
> > > On Mon, Apr 25, 2005 at 09:45:56PM +0200, Bernhard Ege wrote:
> > >
> > >
> > > > Bernhard Ege wrote:
> > > >
> > > >
> > > > > I am trying to copy an 800MB file from my pc to my laptop. The pc has cygwin and cygwins sshd running and from the laptop I use the scp command to copy the file.
> > > > >
> > > > > Result: 190kB/s with low (1-5%) cpu usage on both machines.
> > > > >
> > > > > Expected result: at least 2-3MBps with somewhat higher cpuusage on both machines.
> > > >
> > > > Doh. I completely forgot that in my .ssh/config I had ssh call a script using the ProxyCommand. That command is a shell script that detects which network I am connected to (company or my own) and uses a tunneling ssh command to the destination if on the company network or a direct connection using "connect" if on my own network.
> > > >
> > > > The connect command is a fairly simple program that redirects stdin/stdout to a host:port. This way, I can either use a ssh to host:port if tunneling is required to connect to host:port if not.
> > > >
> > > > The problem is that cygwin has a very low throughput using the script with the connect command. I guess it could be related to the slow pipe problem mentioned earlier (but was supposedly fixed).
> > > >
> > > > I have tried without the ProxyCommand and the speed returned to an acceptable 1.5MBps.
> > > >
> > > > I hope Cygwin can be fixed so the speed returns to normal :-)
> > >
> > > Yep.  We'll get right on fixing that problem where fairly simple
> > > programs which redirect stdin/stdout to a host:port cause reversions in
> > > behavior of the "slow pipe" problem which was supposedly fixed.
> >
> > It was either that kind of reply or some other reply indicating what additional info is required. Obviously I would submit the required information, but I don't know what would be relevant.
> >
> > Anyway, here is the link to connect.c:
> >
> > http://www.taiyo.co.jp/~gotoh/ssh/connect.html
>
>
> > And just create script like ssh-connect.sh:
> >
> > #!/bin/bash
> > connect $1 $2
> >
> > And in .ssh/config add the script to a destination:
> >
> > ProxyCommand ssh-connect.sh %h %p
> >
> >
> > scp'in from the host which is reached through connect has its speed severely reduced.
> >
> > It may be the problem with connect itself, but I am not sure how to test that.
>
>
>
> Seems like something is going on with the compiled version of connect that
> you pointed to.  If I try what you mentioned above with the precompiled 
> version, it is dreadfully slow for me as well (at least several orders of 
> magnitude slower).  Building a local one from source with Cygwin was allot 
> better, though still slower than without it (by 4 to 5 times).  You might 
> want to try a local build if you need to rely on connect.  It's allot better
> than the alternative.

The one I tried _was_ the freshly compiled (by my self) version. I did 
not try the precompiled version. The slowdown you experience (4-5 times) 
approximately matches the slowdown I see as well.

Connect.c is probably not quite cygwin friendly yet (or the other way 
around), but I haven't found any alternative (though I haven't looked 
very hard, only mildly hard ;-).

However, the job connect has to do seems simple enough:

Establish connection in host:port, create 2 threads (or 1 more), one 
thread reading from standard in and output on socket out and one thread 
to read from socket and write to standard out. connect handles a few 
protocols, but for this to work, data should just be passed right 
through, i.e. no protocol handling necessary.

Bernhard

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/