This is the mail archive of the cygwin mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: sshd and scp/sftp: slow throughput on windows machines


At 22:52 2006-03-18 +0100, Max Stein wrote:
>> 1. Is it possible to increase the bandwith by having the client aggregate
>> multiple sessions through a single pipe?
>
>Could you please give me some advice how this can be achieved? I am not an 
>SSH guru yet.

Unfortunately neither am I. It was an idea derived from a report on the
stunnel 
program:
<http://ftp.surfnet.nl/networking/stunnel/>
that tunnels through SSL and according to the report can do such aggregation.
(I don't know an english version of this report so I'll refrain from
providing a 
link to that.)
Since neither the CPU nor the network bandwidth is fully used in your case
it would seem at least theoretically possible that the same could be done
with 
transport over SSH. I formulated it as a question because I am not absolutely 
sure and don't know the details myself.

<snip>
>> W2k and XP have easy to configure PPTP clients.
>> (See also W2003 RAS.)
>
>Why should a point to point tunnel improve the performance? Using Linux on 
>the client and server machines I achieved a throughput of 10.8 MB/s whereas 
>the theoretical maximum on a 100 MBit/s ethernet network would be 12.5 MB/s.
>
>There must be another way. Why is the Linux implementation of SSH able to 
>provide a much better throughput for scp/sftp
>than cygwin's implementation running on the same hardware? It is not a 
>problem of the Windows operating system because usual FTP tranfer yields 
>simalar fast throughput of 10-11 MB/s like SSH running on Linux.

Ah, so your first MB was Megabit and the others were MegaByte...
To prevent any more misunderstandings: Are you talking about the 
Windows FTP or the Cygwin FTP here?
Anyway, it seems not too far fetched to assume that anything that runs 
directly on a native (i.e. Windows or Linux) OS would outperform a similar
thing running through an emulation layer (Cygwin).

Arend-Jan Westhoff

--
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/


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]