phc-discussions - Another dumb idea: compute hardened KDFs

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-next>] [day] [month] [year] [list]

Message-ID: <CAOLP8p7kEAyYwzb5942sa988pWa9F424fvdb3qNwcLshC=z_rw@mail.gmail.com>
Date: Sun, 12 Jan 2014 17:50:26 -0500
From: Bill Cox <waywardgeek@...il.com>
To: discussions@...sword-hashing.net
Subject: Another dumb idea: compute hardened KDFs

Alexander had yet another great idea, and I'm pretty excited about it.
 It may be an old idea, but it's the first I've heard of it.

Multipliers are hand optimized computing engines that go just as fast
on my CPU as on a high-end ASIC (except for liquid nitrogen cooled
ASICs... yeah, they exist).  A 32x32 -> 32 multiply takes 3 cycles, or
almost 1ns on my development machine.  By having a single multiply
operation in the inner loop that computes hash data along with one or
two 1 cycle operations, we can have the multiply delay dominate the
runtime.

An ASIC attacker can speed up the other operations as if they didn't
even exist, but the multiply is hardened, and there's nothing he can
do about it.  To fill the memory bandwidth pipe, we have to run
multiple multiplier loops in parallel, either through instruction
level parallelism or on multiple threads.  My initial tests shows that
we can run multiple threads until the memory bandwidth limit is over
half full.

This way, we harden both the time (multiplies) and cost (memory)
dimensions of our time*cost goal to within 2X of optimal for a given
computer.

What do you think?  Old idea, or is everyone doing this, or for some
reason not?  I can't believe I was giving Alexander grief for
bothering to try to slow down an ASIC attacker by doing a lot of ALU
operations.  Multiply rules.

Bill