phc-discussions - Re: [PHC] A review per day

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]

Message-ID: <540B47B3.2070206@ciphershed.org>
Date: Sat, 06 Sep 2014 13:43:15 -0400
From: Bill Cox <waywardgeek@...hershed.org>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] A review per day - Lyra2

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

To finish some benchmarks, here's Lyra2 hashing 1GiB with one
wandering pass (t_cost == 1):

PHC> time ./phs-lyra 1 174762
Allocated 1073737728 bytes

4b ac 00 17 af 96 a3 66
f2 1a 6a 03 92 6a 9a 23
54 11 10 f0 1a bd c6 56
ea 09 e3 29 c0 00 b7 cb      32 (octets)

real	0m0.701s
user	0m0.629s
sys	0m0.072s

And here's 4MiB:

PHC> time ./phs-lyra 1 682
Allocated 4190208 bytes

2e fe 7c d3 08 9b 48 f0
fc 2a ae 8b f4 d7 16 2a
f1 80 f7 6a 98 22 78 fe
0e 24 6e aa d6 3a 93 d9      32 (octets)

real	0m0.004s
user	0m0.003s
sys	0m0.001s

For comparison, I just measured hashing 1GiB on one thread with TwoCats:

time ./phs-twocats 0 20

95 70 17 bc 36 38 29 9b
31 ef 6c 22 33 63 c5 b1
ad d1 15 c8 e9 a0 fa 8c
49 11 f8 84 84 5f 05 40      32 (octets)

real	0m0.240s
user	0m0.179s
sys	0m0.061s

This difference is speed is mostly due to Lyra2 doing more reads and
writes than TwoCats, and somewhat due to latency in the Blake2b hash
function.  Lyra2 has the 3rd best single-threaded performance I've
measured after TwoCats and Yescrypt.

It's not actually this runtime that makes the most difference, since
we have extra cores that could help out and lower these numbers.  It
is the extra reads and writes to the same memory location
significantly lowers Lyra2's memory*time defense.  For an ASIC attack
that allocates all the memory up-front, TwoCats would have what looks
like a 3X-ish memory*time advantage.  That means my ASIC attack will
probably need only 1/3 the memory of a TwoCats or Yescrypt attack.

And... Lyra2 passes the Dieharder tests :-P  Don't everyone be shocked
at once!  I ran the internal memory at the end of hashing and only one
wondering pass.

Final thoughts... Lyra2 is solid.  I do not like the extra reading and
writing that Lyra2 does to win in the TMTO category.  I think it loses
too much in the time*memory defense category as a result.  This is the
primary reason I put Lyra2 3rd on my list of Scrypt replacement
contenders.  However, if the judges feel TwoCats and Yescrypt are too
complex, Lyra2 is still a solid entry.  I would not feel upset about
Lyra2 being chosen over TwoCats and Yescrypt.

Bill
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
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=r6WR
-----END PGP SIGNATURE-----