Message-ID: <53FFBD82.3070404@ciphershed.org>
Date: Thu, 28 Aug 2014 19:38:42 -0400
From: Bill Cox <waywardgeek@...hershed.org>
To: discussions@...sword-hashing.net
Subject: A review per day - TwoCats

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

How can I objectively analyze TwoCats?  It's one of the coolest things
I've ever written.  If it weren't for Alexander being in the
competition, I honestly feel TwoCats would be the most deserving
successor to Scrypt.  I'm biased in that view, but of course I should
believe this!  If I had thought of way to improve TwoCats, I would
have!  It is very close to being the best algorithm I can come up with
for this problem.  Not only that, but I had a bunch of guys on this
list helping me discover TwoCat's limitations way before the
submission deadline.  It's a result of my single-minded focus on my
hobbies, and great minds here.

All that said, Solar Designer still out-coded me.

The biggest thing I do not like about TwoCats is that I claim to be
it's author, when most of it's ideas are from Alexander (Yescrypt) and
Christian (Catena).  This feels like a lie.  There is waaay too much
in common with both Yescrypt and Catena, and the small portion of
TwoCats which is in neither Yescrypt or Catena is not interesting
enough to win anything.  There's still enough for me to be proud of,
though.

As for other limitations, I think they mostly come from my focus on
password hashing for two applications: ssh private keys and TrueCrypt.
 Both should run for about 1 second, using as much memory as possible.
 At least in the ssh case, it is unlikely that I can convince the code
maintainers to allow me to use all the CPUs on the machine.
Therefore, I decided to hash memory as fast as possible.  In
single-thread mode, TwoCats I believe is the world's fastest hashing
algorithm.  I also feel it is secure, because it just doesn't matter
what you fill memory with, so long as an attacker has to do the same.

One thing I screwed up was how to use t_cost.  For TrueCrypt and ssh
key derivation, the obvious limit is time, not memory.  Therefore,
t_cost should always be set to minimum.  I intend to use TwoCats
always with t_cost set to 1.  What I used t_cost for was an L1 cache
bandwidth hardening parameter.  Ideally, you should set t_cost as high
as you can without having hashing slowing down significantly.  At that
point, you're maxing out L1 cache bandwidth in addition to whatever
else is limiting speed.  That's the best use I could come up with for
a t_cost parameter.

Both Yescrypt and Lyra2 use t_cost for improving TMTO defense.  Duh...
that didn't occur to me!  Why would I use t_cost to *slow down*
hashing?!?  That makes my poor ssh key *less* secure!  Of course,
there are good reasons for slowing down hashing and using *less*
memory, but going slow ins't an area where I tend to innovate...

That's all I have for now!  Please feel free to dump on TwoCats!  It's
my turn :-)

Bill
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
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=7Ki9
-----END PGP SIGNATURE-----

Powered by blists - more mailing lists