Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Wed, 6 Jul 2011 19:03:35 +0200
From: Łukasz Odzioba <>
Subject: Re: Lukas's Status Report - #8 of 15

> That's a curious chart.  Am I correct that with your current code the
> speed is limited by memory bandwidth and/or latency?  This would explain
> the speed reduction with longer passwords, with each and every length
> increase.
Yes, your right.
I've added to wiki similar charts for two other patches, because
comparing patches by pure c/s hides some information for example:
-when we change test vectors, what was often in my case
-when we do optimization which works only for shorter or longer passwords
-other things which cannot be "captured" by those charts

On sha256crypt chart we see performance drops when passlen is grater
than 6. It's because we need to call sha256 compression function twice
(hashed datasize overflows one 512bit block) for every iteration.
Sha512crypt is almost constant regardless of passlen. Md5-crypt calls
md5 only once for every iteration up to passlen equal 15.

>> -Created Sha256-crypt cuda revision 2 (+14% c/s)
>> -Finished Sha512-crypt cuda revision 0 (4794 c/s)
> Is this for rounds=5000 or for another number (maybe a mix of them)?
It was rounds=5k and passlen=10 for both of them.

> Apparently, you haven't done that yet?  I see no SHA-512 stuff there.
> Yes, please upload ASAP, with proper license in the source files.
Sorry I've made a backup and went to bed. All patches are already on wiki.
I've got two additional questions:
-does patches should contain "This file is part of John the Ripper
password cracker." in the license?
-is it possible to delete files from wiki?


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.