Re: [Tails-dev] performance test: randomsound vs haveged

adrelanos@??? wrote (20 Dec 2012 16:14:46 GMT) :
> cat /proc/sys/kernel/random/poolsize is always 4096. Unusable for
> benchmarking.

Sure.

> cat /proc/sys/kernel/random/entropy_avail numbers always change.
> Depends on hardware (real or virtual). Also depending on which
> entropy related packages are installed they change even more.

Full ACK.

> I didn't check the kernel sources either, but this is backed up by
> practical tests and instructions on how to benchmark it everywhere.
> There is a consensus.

Looks like there's a big misunderstanding about my intent. Sorry for
not being clear enough. Maybe "capped by" isn't proper English :)

I was suggesting that, given the default poolsize (that is, in my
understanding, the maximum amount of entropy in the pool) is 4096,
comparing measurement results that are around 4k has limited meaning.
It shows that each of the tools that fill the pool fill it entirely
(both running alone, and combined).

I'm not saying these results are useless (they are probably all we
need to know in practice), I'm just mentioning it because you wanted
to check if running haveged and randomsound at the same time results
in even more entropy available... and without any experiment, one
could tell that ($max + $max) can't possibly be greater than $max.
See what I mean? :)

Thanks anyway!