Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 10 Nov 2015 09:52:16 -0800
From:      John-Mark Gurney <jmg@funkthat.com>
To:        Dag-Erling =?iso-8859-1?Q?Sm=F8rgrav?= <des@des.no>
Cc:        freebsd-current@freebsd.org, freebsd-security@freebsd.org
Subject:   Re: OpenSSH HPN
Message-ID:  <20151110175216.GN65715@funkthat.com>
In-Reply-To: <86io5a9ome.fsf@desk.des.no>
References:  <86io5a9ome.fsf@desk.des.no>

next in thread | previous in thread | raw e-mail | index | archive | help
Dag-Erling Smrgrav wrote this message on Tue, Nov 10, 2015 at 10:42 +0100:
> Therefore, I would like to remove the HPN patches from base and refer
> anyone who really needs them to the openssh-portable port, which has
> them as a default option.  I would also like to remove the NONE cipher
> patch, which is also available in the port (off by default, just like in
> base).

My vote is to remove the HPN patches.  First, the NONE cipher made more
sense back when we didn't have AES-NI widely available, and you were
seriously limited by it's performance.  Now we have both aes-gcm and
chacha-poly which it's performance should be more than acceptable for
today's uses (i.e. cipher performance is 2GB/sec+).

Second, I did some testing recently due to a thread on -net, and I
found no significant (not run statistically though) difference in
performance between in HEAD ssh and OpenSSH 7.1p1.  I started a wiki
page to talk about this:
https://wiki.freebsd.org/SSHPerf

Feel free to add to the page any more info.

There are other apparent issues w/ ssh that keeps it's performance low
on high latency links, but I haven't spend the time to figure out what
they are, but in my testing HPN did not increase performance to make
use of the fat but high latency link.

So, if it's not increasing performance and making us fall behind, why
bother with the trouble of keeping the patch?

If someone is willing to spend the time doing benchmarks, and prove
that the HPN patches do make a difference, I'm willing to work with
them to figure out why my tests didn't work and change my vote.  I
also believe that the defaults should be enough, if you have to tune
or enable features, then you can install from ports or compile yourself.

-- 
  John-Mark Gurney				Voice: +1 415 225 5579

     "All that I will do, has been done, All that I have, has not."



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20151110175216.GN65715>