1
$\begingroup$

I was reading the NIST 8114 draft and there's an example of a profile, Sample Profile #2, that shows that 20ns (50MHz) for a block completion for encryption on a CANBUS. The bus is serial and maxes out at 1us (1MHz). Is there some standard somewhere that specifies a 20ns completion time for a block cipher? I cannot figure out, from a hardware standpoint, where that 20ns came from.

$\endgroup$
  • 1
    $\begingroup$ There is such a thing as latency. You may not just want encryption but also MAC. And you may also want to do something with the data you receive. So such timings do seem about right, but I don't know that standard, so that's just educated guesswork that might give you some insight. $\endgroup$ – Maarten Bodewes Oct 4 '16 at 19:19
  • 1
    $\begingroup$ 20ns per block * 64-bit block = 3.2Gb/s, which is pretty damn quick for "lightweight" crypto over something like CANBUS, but is however a reasonable speed for something like CBC-MAC over the type of expected data $\endgroup$ – Richie Frame Oct 5 '16 at 6:27
  • 1
    $\begingroup$ I agree than 20ns latency has no justification for "command validation on a Controller Area Network (CAN) bus", where the timescale is much larger. On the other hand, the quoted sentence occurs as a possible application of hypothetical Sample Profile #2, is not given as justification of the low latency, and there's nothing to tell that the numbers in this sample profile make any sense. $\endgroup$ – fgrieu Oct 5 '16 at 7:11
  • 1
    $\begingroup$ Perhaps they were just using performance numbers from an off the shelf product as a baseline. Samsung Exynos 5433 can push 10.6Gb/s of AES on a single ARM Cortex A57 core @ 1.9GHz, which is 13ns per 128-bit block (presumably for 128-bit key, or approx 17ns per block for 256-bit) $\endgroup$ – Richie Frame Oct 6 '16 at 1:06
  • $\begingroup$ I'm going to the NIST 8114 workshop and I will try to figure out why that number was picked. I thought that the best way to know was to go, and perhaps I can answer my own question. None of the "automotive" MCUs on digikey can come anywhere near these figures and the CANBUS is the primary constraint anyway, but as Richie mentioned, it makes sense from the CPU side. Maybe there's someone out there using CANBUS protocol on 10Gb/s Ethernet. $\endgroup$ – b degnan Oct 6 '16 at 16:04
5
$\begingroup$

The authors of the NIST 8114 draft said that the times were arbitrary for the purpose of generating discussion. They did, indeed, generate discussion.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.