HPH-SSH code incorporation

[Chris Rapier]

Darren Tucker wrote:

> alan.cl.wong at nokia.com wrote:
> 
>>	We are deciding on to either entirely use HPH code to our
>>OpenSSH or not for systems globally and I come down to a question.
> 
> 
> I assume you mean HPN?
> http://www.psc.edu/networking/projects/hpn-ssh/
> 
> 
>>If HPH-SSH code is such an improvement to network performance for
>>OpenSSH, then why has it not been incorporated to the OpenSSH code? Is
>>there a reason? Is it because there are problems with the HPH-SSH code?
> 
> 
> (speaking for myself) Short answer: lack of time to review and 
> difficulty testing the patch and/or variants thereof.
> 
> I did a review a of the (current at the time) patch while back:
> http://marc.theaimsgroup.com/?l=openssh-unix-dev&m=112316226728255
> As far as I can see, none of those items have been addressed.

We haven't actually had time to review your proposed changes and decide if we 
want to incorporate them. Its not that we think that they are bad ideas. I 
personally have no real problem with incorporating them once they are tested 
out but we've moved onto a different aspect of this project and this didn't 
just move to the back burner but fell off the stove. I'll pick it up, brush it 
off, and see what we find out.

>>Since the code has been out for a time and still has not been
>>incorporated to the OpenSSH code. It creates doubt of there must be a
>>reason why the OpenSSH group has not incorporated the patch. There are
>>concerns of code scrutiny and review if it is not incorporated. OpenSSH
>>has now become not just an application but a medium for data transfer
>>and that means network performance is very critical. If network
>>performance can be increased without decrease in security then it would
>>make OpenSSH even more attractive.
>>
>>So the question is is there a reason why the OpenSSH group decided not
>>to incorporate the code into theirs?
> 
> 
> It basically needs to be split up, each piece tested individually and, 
> if possible, simplified.  We've worked the the HPN folk this way in the 
> past (eg bugzilla #896).

My impression of that was that 896 was a bug. Addmittedly, the only way it 
could have been triggered was with our patch. Mike Stevens has more detail on 
that.

> The testing part is difficult for folks without a transcontinental ATM 
> link or similar handy.  (I did some testing with a software solution to 
> add latency but spent more time debugging the test rig than ssh.)

Currently we have test sites in switzerland, Pittsburgh, Pennsylvania, and 
Boulder, Colorado. We might be able to provde access to those sites (I need to 
review our policy on this) in some limited manner. If I can then you'd be on 
rather hjigh BDP links. Not GigE but at least several hundred Mb.