Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 13 May 2012 22:27:13 +0530
From: Dhiru Kholia <>
Subject: Re: SSH thread-safety

On Sun, May 13, 2012 at 9:47 PM, Solar Designer <> wrote:
> Dhiru -
> On Sun, May 13, 2012 at 08:46:38PM +0530, Dhiru Kholia wrote:
>> I have added a self-test key with key length of 2048 bits.
> Oh.  So it's one 2048-bit and one 1024-bit key that are being used for
> the benchmark now, right?  Perhaps this should be reflected in the
> comment.  Right now, BENCHMARK_COMMENT is set to empty string, but
> perhaps it should be " one 2048-bit and one 1024-bit key".  And while
> we're at it, perhaps FORMAT_NAME should be uppercase "SSH".  Either of
> these changes will break relbench of this format against older versions,
> but that's as desired because we've changed what's being benchmarked.

I have implemented these changes.

> I've tried changing BENCHMARK_LENGTH from -1001 to -1000 or 0 to get
> separate benchmarks for many vs. one salt.  Somehow the results are
> almost the same, even though with two test keys of different size I
> expected them to differ - aren't we using just one of the tests for the
> single-salt benchmark?  Any explanation why I did not see a significant
> change in speed in this experiment?  Also, any explanation why the
> benchmark speed is now very similar to actual cracking speed for
> 2048-bit keys rather than somewhere in the middle between that for 2048-
> and 1024-bit keys if both are seen in tests[]?

Switching the order of self-test vectors gives us the old speeds back.

For some reason, JtR is using only the first self-test vector (it uses
the second entry only one time, I added a printf in set_salt to verify
this).  This applies even if BENCHMARK_LENGTH is 0.


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.