[Beowulf] Re: High Performance SSH/SCP

[Benjamin Bennett]

Mark Hahn wrote:
[snip]
> the switch-to-none feature is a good one.  I'm a bit skeptical about
> the buffering changes, since you can accomplish the same thing with 
> sysctls.  (has anyone ever experienced a real case where too-large 
> sysctl net memory settings caused problems?  obviously, attempting to do 
> long-fat-pipe transfers to a heavily used web server might be a problem, 
> since the latter wants tight controls on sock mem use.)
> 
> the really cool thing would be if you could associate a default setting 
> for socket buffers with a _route_.  heck, a round-port combo.
> it seems crazy for apps to be messing with these issues.

The SSH protocol has its own, per channel, flow control on top of TCP to 
multiplex multiple SSH channels over a single TCP stream.

The stock OpenSSH code has static window sizes, IIRC 64KiB in <= 4.6, 
and 1-2MiB in 4.7

HPN-SSH dynamic windows does not adjust the TCP buffer size, but these 
SSH channel windows so that they're not a limiting factor when you have 
an auto-tuning kernel, or have manually adjusted TCP buffer size above 
these static values.

This is much better described in our paper [1] and the presentation of 
that paper [2].

A colleague just pointed me to this thread, I'll try to keep an eye on 
it if there are any questions, or feel free to contact hpn-ssh at psc.edu

--ben


[1] http://www.psc.edu/networking/projects/hpn-ssh/papers/a14-rapier.pdf
[2] 
http://www.psc.edu/networking/projects/hpn-ssh/papers/MG08-rapier-bennett.ppt