Bench Testing

[mk01]

@IriDium,

prime number computation is the most CPU only test ever.
it is from 99.9% dependent on CPU frequency and CPU internal architecture (nr of math units, pipes in that case so CPU can do X multiplications in one clock tick even if being single core).

that's why 5000 primes it takes 90s on CortexA9(1ghz) core, 4s on i5(3.4ghz) and only 3s on i7(being only at 2ghz).
there is no involvement of memory, not even internal processor caches.

but try that :
sysbench --test=threads run

on 3.17 on RPI and 3.14 cubox, 3.10 cubox, 3.0.35 cubox. this test can make use of effective scheduler/locks handling.
and RPI on 3.17 gives ~~ as 3.14 cubox (with lower freq and inferior CPU design).

so context switching and wasted time on internal management in kernel
translated to user feel - responsiveness.

we used to use 2.4 kernels... in last century. don't remember it much but remember 3.0.35 on cubox. pain (but most Androids run it).