phc-discussions - Re: [PHC] yescrypt throughput vs. PWXrounds

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 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: Fri, 3 Apr 2015 23:11:18 +0300
From: Solar Designer <solar@...nwall.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] yescrypt throughput vs. PWXrounds

On Fri, Apr 03, 2015 at 09:05:59PM +0300, Solar Designer wrote:
> rounds  2 MB            128 MB          2 MB + 2 GB ROM
> 6       2772 / 511      30 / 7          2592 / 486
> 4       3653 / 691      32 / 9          3269 / 647
> 2       5340 / 1077     33 / 13         4288 / 974
> 1       6454 / 1451     33 / 15         4760 / 1255
[...]
> [...] claim that yescrypt achieves bcrypt-like frequency(*) of its
> S-box lookups and thus is at least as GPU-unfriendly as bcrypt even at
> the lowest m_cost settings.  Would being no more than 2x worse than
> bcrypt still be OK?  I'm not sure.  I would be uncomfortable about that,
> even though bcrypt isn't one of the PHC finalists. ;-)
> 
> (*) Also considered are parallelism of the S-box lookups and total size
> of the S-boxes.

One way to compensate for the reduced number of rounds in terms of
bcrypt-like anti-GPU is to double the size of the S-boxes.  With the
S-boxes doubled from 8 KB to 16 KB, and PWXrounds = 2, I get 4212
hashes/second instead of the 5340 hashes/second in the table above.
This gives 2772*6 / (4212*2) = 1.97 reduction in compute hardening, but
2772*6 / (4212*2*2) = 0.99 almost no change in bcrypt-like anti-GPU,
both compared to the current defaults of 8 KB and 6 rounds.

So it is possible to retain bcrypt-like anti-GPU at lower PWXrounds, but
I am not happy about the almost 2x reduction in compute hardening.  And
the speedup achieved with the reduced PWXrounds is less, because some of
it is lost to L1 cache misses on the S-box lookups.  16 KB is a bit too
large since we have other data as well (current block being processed).
IIRC, the impact on Intel CPUs is less, though.

At 128 MB, PWXrounds = 2 with 16 KB S-boxes gives 31 / 11, as compared
to 33 / 13 in the table above (for 8 KB S-boxes).

Alexander