Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 26 Apr 2015 09:48:15 +0800
From: Lei Zhang <zhanglei.april@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: [GSoC] JtR SIMD support enhancements


> On Apr 26, 2015, at 1:39 AM, Solar Designer <solar@...nwall.com> wrote:
> 
> Looks reasonable now.  Just a bit slower than expected:
> 
> 5888/6339 = 93%
> 236/240 = 98%
> 
> Is the overhead of offloading so high that we have a 5% performance hit
> even at these low c/s rates?  I guess the initial S-boxes get
> transferred from host every time?  

Yes, it's transferred from host to MIC every time. It's even transferred back from MIC to host, which is the default behavior of offloading. I tried tweaking with the pragma options to eliminate unnecessary transfers, but gained no significant improvement.

Code:
-----------------------------------------------------
#pragma offload target(mic) in(salt:length(1)) in(BF_magic_w) in(BF_init_state) in(BF_exp_key) in(BF_init_key) out(BF_out)
...
-----------------------------------------------------

Output:
-----------------------------------------------------
[zhanglei@...ter src]$ ../run/john --test --format=bcrypt
Will run 12 OpenMP threads
Benchmarking: bcrypt ("$2a$05", 32 iterations) [Blowfish 32/64 X2]... DONE
Raw:    5900 c/s real, 5911 c/s virtual
-----------------------------------------------------

> Do you have access to any machines with more than one MIC card, to see
> how offload would behave there?

Sorry, I don't.

> Anyway, did you commit this code to some branch?  Non-default, I guess,
> since it's highly experimental and mostly breaks things?  We should
> preserve it somewhere for future experiments.

The modifications I made to the code is a bit nasty for the moment. It'll need more polishing before getting committed.


Lei

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ