Re: Why SFTP performance sucks, and how to fix it

[Nicolas Williams <Nicolas.Williams%sun.com@localhost>]

On Tue, Jul 08, 2003 at 06:05:25PM +0200, Markus Friedl wrote:
> On Tue, Jul 08, 2003 at 08:46:10AM -0700, Nicolas Williams wrote:
> > Consider an implementation such as OpenSSH where the SECSH protocol and
> > the SFTP protocol are implemented in separate processes communicating
> > over IPC.  With such an implementation one might, in some situations,
> > end up having a bottleneck in the IPC connection, rather than in the
> > client-server TCP connection, and under such circumstances one might
> > like the SECSH channel flow control mechanism to work as advertised.
> 
> feel free to send patches....

Before we talk about patches I'd like to get a few more details
straightened out.

OpenSSH has a per-channel max window size that controls the max value
for the actual window size.

I think a way slow-start (and slow-stop) could be implemented would be
to adjust the per-channel max window size according to the status of the
per-channel data sink when data is received:

 - Slow-start

   When data is received on a channel and the channel's sink is
   writable, increase the channel's max window size by some amount.

 - Slow-stop

   When data is received on a channel and the channel's sink is not
   writable, reduce the channel's max window size by some amount, but
   don't allow the channel's max window size to become lower than the
   actual window size.

The question is: what should this "some amount" be?  It could a
constant, or the size of the data received that triggered the max window
size adjustment.  Or maybe the max window size adjustments algorithm
could be more aggressive so that slow-starts/stops aren't that slow...

Cheers,

Nico
--