Re: RSA Key generation time

[Mithun P <getmithunp@xxxxxxxxx>]

Hi,

I tried the same key generation on the default linux port from freescale on the same board and i am getting an average of 20 seconds with the same board.

Do you think that there is such a huge performance margin with OS.

The only other difference that i can see is that on the VX works port of openssl, i am using a compiler with is old and optimized for e-500 core while P1010 uses a e-500V2 core.

Also on the VXworks port of OpenSSL, i have enabled FIPS whereas in the linux port of openssl FIPS is disabled.

Regards

Mithun

On Sat, Jan 28, 2017 at 4:58 PM, Frank Migge <fm@xxxxxxxxxxxx> wrote:

Hi Mithun,

>> I have a embedded board P1010 RDB  running openssl on VXWORKS 5.4 .

>> I am generating RSA 2048 and 3072 bit key pairs.

>> I am providing entropy to openssl by using RAND_seed from a HW RNG.

>> My average generation time for RSA 2048 key pair is 2 Minutes and  3072 is 8 minutes.

I noticed embedded board key generation times vary by OS and OpenSSL version after converting a Altera Atlas FPGA SoC HPS from original 2013 Yocto Linux to latest Ubuntu. Under the old Yocto, key generation occasionally took up to 2 minutes. Same board under Ubuntu 16.04, 2048 RSA keys take consistently 2-5 seconds, while 3072 keys need around 8-16 seconds. Even running the system single core, the numbers don't change (on a low utilized system, using OS built-in /dev/urandom).

While I am on a different CPU and OS (32bit ARM v7  900Mhz dual core, 1GB 400Mhz RAM), your e500 PowerPC can't be to far behind. Your numbers seem to be off by a magnitude. You mentioned using a external HW RNG, could that be it?

Cheers,
Frank

Jakob Bohm

Wednesday, January 25, 2017 1:10 AM

I'm afraid you will have to look at the OpenSSL source code, I haven't
paid much attention to that CPU recently.




Enjoy

Jakob

Mithun P

Monday, January 23, 2017 4:09 PM

Hi Jakob,

Can you please give me some reference/example of bignum optimization which I can check on powerpc architectures.

Is this any specific instruction set addition? or something more generic?

Thanks & Regards

Mithun

Jakob Bohm

Wednesday, January 18, 2017 1:08 AM

I believe this is a CPU intensive operation (if VxWorks can do
this, try observing the CPU load during).

Potential improvements:

1. Check if the CPU specific bignum optimizations for your CPU
  variant have been enabled via the libcrypto CPU detection code
  (for example, there are optimizations for different ARM cortex
  variants).
2. Faster CPU (expensive obviously).
3. Do the generation in the background before the keypair is
  needed, at a time when the extra CPU load is less of a problem.


Enjoy

Jakob

Mithun P

Tuesday, January 17, 2017 3:44 PM

Hi 

I have a embedded board P1010 RDB  running openssl on VXWORKS 5.4 .

I am generating RSA 2048 and 3072 bit key pairs.

I am providing entropy to openssl by using RAND_seed from a HW RNG.

My average generation time for RSA 2048 key pair is 2 Minutes and  3072 is 8 minutes.

Is there a way to reduce the generation time?

Regards

Mithun

--
openssl-users mailing list
To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users

-- 
openssl-users mailing list
To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users