[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+EaD-ZYmVoFo_uV+TnCs-FwGijsSq9SJcjZNa9PsTYYHrzXvg@mail.gmail.com>
Date: Sat, 13 Jul 2013 21:45:29 +0200
From: Katja Malvoni <kmalvoni@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: Parallella: bcrypt
On Sat, Jul 13, 2013 at 9:09 PM, Solar Designer <solar@...nwall.com> wrote:
> [..]
Oh, these times are much better than what you reported before. IIRC,
> previously we were unable to get much better than 30 ms - and now you're
> saying you're down to 20 ms already, which means almost 50 c/s. Your
> 20.7 ms should give us 773 c/s on 16-core Epiphany, which is roughly the
> speed of one modern x86 CPU core - but at much lower power usage. The
> 64-core Epiphany is thus comparable to Core 2 Quad from a few years ago -
> but again, at much lower power usage.
I haven't realised that I never reported those, I'm sorry about that. This
improvement happened when I changed Epiphany code so that it can be
integrated in JtR. Most of it comes from returning result in BF_binary
instead of returning much bigger char array.
> You just need to implement this into a working JtR format now, and do so
> in a way that would let the implementation scale at least to 64 cores.
When my current code starts working, this will be easy, it'll require only
few changes of constants.
> [...]
>
> Using these numbers, I get:
>
> 2404/20.7/128 = 91% efficiency
> 8717/20.7/512 = 82% efficiency
>
> These are somewhat inconsistent - I would have expected to see almost
> the same efficiency percentage from these two, since cost 12 is already
> high enough.
>
For cost 12 correct number is 2182, 2404 is execution time with mine
non-optimal assembly code.
In that case:
2182/20.7/128 = 82,4 %
[...]
> Do you feel your work on getting JtR integration working correctly is
> still pending Yaniv's help?
>
> Alexander
>
I do, I ran out of ideas and I can't figure out what goes wrong on Epiphany
side. And all things that I've tried in past few days result in same or
worse behaviour.
Katja
Content of type "text/html" skipped
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
