Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 31 Dec 2012 10:24:46 +0100
From: magnum <>
Subject: Re: Gladman vs Lukas (was: Formats dmg, encfs and strip crash on longer passwords)

On 31 Dec, 2012, at 7:30 , magnum <> wrote:
> On 31 Dec, 2012, at 5:22 , magnum <> wrote:
>> On 31 Dec, 2012, at 5:03 , Dhiru Kholia <> wrote:
>>> On Monday 31 December 2012 09:11 AM, magnum wrote:
>>>> I see we have some formats that use Gladman's derive_key() instead. This is slower. I tried changing ODF to keychain.h and pbkdf2() and got a 60% boost but I'm not sure it supports all variants (if there are any?) so I did not commit that. Gladman's function has one more parameter and I'm not sure if it matters. I also tried SXC but got no boost, no idea why. Finally, I tried ZIP but that did not even pass self-test.
>>> 3. SXC and no boost is a mystery.
> I found the reason: SXC only use 16 bytes of output (at least all test vectors do). So while Gladman's code may be slow, Lukas code does twice as many iterations - just to produce longer output than needed. It's nearly as fast despite this! The best solution would be to add that last key length parameter to Lukas' function. This would also let you use his code for ZIP!
> I can give this a shot.

Done. ZIP and SXC got a 2x boost. Also, SXC-opencl got 4x (ignoring the fact it failed self-test before this) and ODF-opencl got 2x. Not that it shows for the OpenCL formats because they lack device tuning, even manual one, so they run on a fraction of the speed they can do.


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.