phc-discussions - Re: [PHC] Re: The best of the best, IMO

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Sun, 13 Apr 2014 20:11:46 -0500 (CDT)
From: Steve Thomas <steve@...tu.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Re: The best of the best, IMO

> On April 12, 2014 at 8:33 PM Bill Cox <waywardgeek@...il.com> wrote:
>
> By Script inspired, I mean algorithms that bust out of cache and do a *lot*
> of memory hashing, like Script.  To do that, an algorithm requires a
> strategy for mitigating cache-miss penalties.  Some entries look to me like
> simplified versions of Script, but those that dropped Script's ability to
> efficiently bust out of cache I put in the Bcrypt-like category.  I think
> it makes more sense to compare them against algorithms like Pufferfish and
> Battcrypt than against large memory capable algorithms like Lyra, Yescript,
> EARWORM, and TwoCats.

Battcrypt should scale linear with memory usage. Since it encrypts blocks
of 2 KiB at a time and the 4 KiB of s-boxes should stay in cache. The only
problem is that Blowfish is slow compared to other algorithms. I think it's
max speed is going to be around 185 MiB/s (3.5 GHz).

Centrifuge would also be able to bust out of cache if they made "out"
larger then once it's done shrink it down. Since this probably will have
similar speeds to RC4 it will be 2.6 times faster than battcrypt because
Blowfish takes 18 clocks/byte* vs 7 clocks/byte**. This is outdated since
both of those were measured on a Pentium. Although Centrifuge doesn't swap
for every byte of output, every block of "out" it swaps based on "t_cost"
then encrypts "out" with AES. I'm guessing with AES-NI it might be a little
slower than RC4.

Really what it comes down to is we need to have optimize versions of them
all, but this is a lot of work. Personally I'm probably just going to wait
until next round when there are fewer to do.

* https://www.schneier.com/blowfish-speed.html
** https://en.wikipedia.org/wiki/RC4

----
Disclosure: I submitted battcrypt and Parallel.