[fedora-arm] 32 bit vs 64 bit for Elliptic Curves

[Peter Robinson]

On Thu, Aug 7, 2014 at 3:50 AM, Robert Moskowitz <rgm at htt-consult.com> wrote:
> There is extensive discussion on the IRTF's crypto list on choosing the next
> Elliptic Curve algorithm  EC25419 may be leading in the discussions.
>
> But the interest here is dealing with lots of ECC calculations on a server
> that is running an SMTP MTA using STARTTLS and DANE a lot. In such a case,
> being able to do lots of EC calcs is important. This gets us to comparing
> the armv7 to the armv8.  I have been told that the armv8 is 64bits, so...

Is that really what the IRTF considers a useful standard representative usecase?

> Is there support for armv8 systems with F21 and of course are there
> affordable armv8 cards?  Or am I getting too ahead of the curve?  ;)

Define "ARMv8"? There's aarch64 support for the ARMv8 architecture as
a secondary architecture (see the daily rawhide reports to this list)
but you can run an ARM 32 bit userspace on those chips.

There's also hardware available but define "affordable"... at the
moment the first gen widely available hardware will set you back about
USD$ 3K

Ultimately ARMv7 / 32 bit isn't going away quickly, ARMv8 is 64 bit,
both have NEON SIMD which provides 128bit SIMD instructions and
there's the ability to use those units to optimise offload of crypto
functions so I suspect NEON is likely the best means of achieving
speed and optimisation use case that will work across both platforms.
ARMv8 has HW optimised crypto extensions but they'll be for existing
cyphers and I've no idea how useful they will be for future standards.

Peter

[1] http://www.arm.com/products/processors/technologies/neon.php