High Performance SSH/SCP - HPN-SSH when?

[Chris Rapier]

I ran some preliminary performance tests on LAN to LAN connections. 
Right now it is just OS X 10.4 to OS X 10.4. I'm not seeing any 
significant performance difference between the various implementations 
regardless of the combination. Anyway, I'm on vacation until Monday but 
when I get back I'll try some linux to cygwin tests to see if I can 
replicate your problem. I'll try some other combinations in the next few 
weeks as well.

Corinna, have you tried using any of the config switches to see fi they 
have an impact on the performance?

Darren, if you have any other suggestions please let me know. I'll also 
try your patch and see what information I can get out of it.

Okay, back to drinking... :)

Corinna Vinschen wrote:
> On Apr 12 22:10, Darren Tucker wrote:
>> On Wed, Apr 12, 2006 at 01:46:32PM +0200, Corinna Vinschen wrote:
>>> Looks like another buffer might be in the way now, but I didn't look
>>> into this deeper so far.
>> Attached is a patch I started to play with that outputs the buffer
>> high water mark for each connection.
> 
> Results for
> 
>   cygwin> scp linux:file .
> 
> 1. Vanilla with 8K buffers:
> 
>   file  100%  118MB   1.3MB/s   01:35    
>   buffer hiwater 1105968
> 
> 2. Vanilla with 85K buffers:
> 
>   file  100%  118MB   5.6MB/s   00:21
>   buffer hiwater 202176
> 
> 3. HPN with 8K buffers:
> 
>   file  100%  118MB   1.0MB/s   02:00    
>   buffer hiwater 1069056
> 
> 4. HPN with 85K buffers:
> 
>   file  100%  118MB   9.9MB/s   00:12    
>   buffer hiwater 1205576
> 
> Results for
> 
>   linux> scp file cygwin:
> 
> 5. Vanilla sshd
> 
>   file  100%  118MB   9.1MB/s   00:13
>   slave buffer hiwater 1085440
> 
> 6. HPN sshd
> 
>   file  100%  118MB   2.0MB/s   00:58
>   slave buffer hiwater 1085440
> 
> It's interesting that the buffer in case 4 is reproducibly only filled
> up to 200K, while in all other cases the buffer is filled up to 1 Meg.
> Is there anything we can learn from that?
> 
>>> Note that all tests are using a vanilla OpenSSH on the Linux side.
>>> I didn't want to add another variable to the test configuration.
>> Be aware that some of the hpn changes only kick in if it sees the other
>> end is also hpn.
> 
> I am aware of that.  The idea is that I have only the Cygwin side under
> control since that's what I'm maintaining.  The generic case is the one
> in which I don't know anything about the remote side and can't influence
> it.
> 
> 
> Corinna
>