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: Thu, 30 Apr 2015 07:57:21 +0300
From: Solar Designer <>
Subject: Re: [PHC] Added multi-threading support to test suite

On Wed, Apr 29, 2015 at 12:15:41PM -0700, Bill Cox wrote:
> Here's a summary of the fastest speeds, when run on my Xeon E5-1650 v2 @
> 3.50GHz.

How many and what kind of memory modules do you have installed?  I've
just checked, and this CPU has 4 memory channels, giving up to 59.7 GB/s
with 4x DDR3-1866.  What does dmidecode say?

We only get:

2^30/10^9*2/0.103 = 20.85 GB/s

yescrypt PWXrounds=2:
2^30/10^9*8/3/0.160 = 17.9 GB/s

yescrypt PWXrounds=6:
2^30/10^9*8/3/0.175 = 16.36 GB/s

2^30/10^9*2/0.151 = 14.22 GB/s

This includes memory allocation overhead, though, which could easily
amount to 30% of the total running time, and possibly way more when
we've sped up the computation with thread.  Those large malloc()'s or
mmap()'s are sometimes surprisingly slow, perhaps because the kernel has
to zeroize those pages before letting us have them.  So maybe we should
actually count like this:

2^30/10^9*3/0.103 = 31.27 GB/s

and so on, for the defender's wasteful memory bandwidth use. :-(

(A case for in-kernel KDF.  Ouch.)


Powered by blists - more mailing lists