[Verse-dev] Verse encryption woes

[Gert Svensson]

Why not ask a local expert Hans Riesel riesel at nada.kth.se?

Best Regards
Gert

Emil Brink wrote:

>Hi.
>
>For the past week and a half, I've been working on integrating the
>code for login-encryption into the Verse codebase. This has involved
>reviewing (and basically rewriting) the core algorithms used in
>generating keys and doing the encryption. I've also rewritten the
>code for big numbers totally, since it turned to match a set of
>requirements that were not correct.
>
>The problem now is that the code seems to be too slow for practical
>application, and I (or even we) am a bit stumped about what to do
>next.
>
>To be specific, it's the prime-testing needed to generate RSA keys
>that is taking too long. Currently, on my lowly 1 GHz laptop, it
>takes about 11 seconds to generate a pair of RSA keys of 512 bits.
>
>Profiling leads me to believe that the most major culprit is the
>division of long integers in v_bignum.c, and this is not very sur-
>prising as the function basically does "long division" the same
>way you do it with pen and paper. I am not a numerics expert, and
>it shows in that code I guess. :/
>
>Does anyone know of a faster algorithm for dividing arbitrary-
>length integers, preferably using a binary representation?
>
>Since the keys should optimally be far longer (2,048 bits was the
>original length, I dialled it down to get results quicker for
>practical reasons), this is an issue that needs to be resolved.
>
>I've suggested to Eskil that perhaps clients (as well as hosts)
>need to buffer encryption keys rather than re-generate them for
>each session, but that doesn't fit very well into the design of
>the API and raises some annoying portability issues.
>
>Comments welcome, as always.
>
>Regards,
>
>/Emil
>
>_______________________________________________
>Verse-dev mailing list
>Verse-dev at projects.blender.org
>http://projects.blender.org/mailman/listinfo/verse-dev
>  
>