Re: [openpgp] SHA-x performance

[Phillip Hallam-Baker <[e-mail]>]

On Wed, Aug 12, 2015 at 9:23 AM, ianG <[e-mail]> wrote:

> On 12/08/2015 03:08 am, Peter Gutmann wrote:
>
>> Werner Koch <[e-mail]> writes:
>>
>> Do you have a suggestion on what CPUs from low to high end to do
>>> benchmarks
>>> so to check which SHA variant is suitable?
>>>
>>
>> It'd be a longish list :-).  I'm also not sure how easy it'll be to get at
>> them, this is all embedded-systems stuff so you can't just SSH into a box
>> and
>> run the tests.  In any case the most common is ARM Cortex M (and a few
>> R), so
>> any Cortex M3 (the example I gave was an STM32 at 180MHz which admittedly
>> is
>> the top end of the range, 100 or 80 Mhz would be another benchmark level,
>> but
>> then they're all ARMv7-M so you can extrapolate from one clock speed to
>> another), then for PowerPC something like an MPC560x at 80 MHz (very
>> common in
>> industrial and automotive), a PIC32MX (MIPS32) at 60 MHz, and then for
>> exotica
>> maybe a NIOS II or MicroBlaze at 100 Mhz.  You don't really need to do
>> dozens
>> of variants since things mostly scale directly with clock speed, and in
>> any
>> case those are reasonably representative clocks, at least for the faster
>> devices
>> where power-saving isn't an issue.
>>
>
>
> To what extent are we accepting the embedded market as "our market" ?
>
> Is this something that already exists in the sense that a lot of them are
> already doing OpenPGP signing for some purpose?  Is the process things like
> signed updates, or are people using OpenPGP to encrypt and/or sign requests
> to the devices?
>
> Or, are we making a stab at the future:  "IoT will need security systems,
> and OpenPGP will be used to supply that, so we'd better make sure it fits
> on the rough template of devices."
>
> To give a ludicrous counter-example, we could also optimise the future
> hash for smartcards.  That's another big market, why not?  I'd say this
> doesn't work mostly because the smartcard or tokens market is mostly
> closed, and systems are typically constructed to be tightly bound to the
> end-user application;  generic systems such as OpenPGP would find it hard.
>
> Where do we draw the line?  Are embedded / IoT inside?


If we are picking algorithm suites then we should look at the cases where
the choice might matter rather than when it does not.

IOT looks set to create a demand for an absolutely minimal cryptographic
suite. One signature algorithm, one exchange algorithm, both on the same
curve, one authenticated encryption mode, one digest/pseudorandom function.

That suite is going to be the one that finds its way into hardware
accelerators and those are in turn the sort of things that are going to be
found as standard cell and on smartcards and such.

So looking at a five year horizon, that is the set that is interesting.