1

I am currently benchmarking the cryptographic throughput of a Banana Pi using the following page as a starting point.

However, I could notice that the dd operation presented doesn't utilize both cores, and read about a related question. So I passed on two instances of dd, but must have missed something because the throughput doesn't increase, and only one one core at a time is used, as shown with htop. The reference is, of course, the benchmark command, only available in the 1.6x branch of cryptsetup, which gives approximately 68% increase in speed with overclock, but roughly 33% increase using the manual method. I strongly suspect the testing method doesn't use both cores.

Would there be a way for dd to use both cores for the same task and report one throughput?

My ultimate goal would be to prove that manual benchmarking process is as reliable for older versions of cryptsetup, which don't have the "benchmark" command, as they are using it, in order to compare it to the Raspberry Pi.

Community
  • 1
  • 1
Cuby
  • 11
  • 1
  • 2
    `dd` is *not* multicore/concurrency aware. The linked answer spawns multiple *different* `dd` processes, and the OS schedules them optimally (eg. on different cores as appropriate). Write the benchmark collector be aware of this fact and add the discrete results as applicable. Also note that `dd` is most certainly *I/O bound* in which case it doesn't matter how hot it makes the CPU, that is just wasted electrons - the disk (I/O) isn't going to give up/take the data any faster. – user2864740 Sep 09 '14 at 04:32
  • In order to force the use of multi-cores with `dd` you will need to use something like `gnu parallel`. See: [**Gnu Parallel**](http://www.gnu.org/s/parallel) – David C. Rankin Sep 09 '14 at 05:25

0 Answers0