combining entropy

Wouter Slegers wouter at
Fri Oct 24 10:23:36 EDT 2008


> If I have N pools of entropy (all same size X) and I pool them
> together with XOR, is that as good as it gets?
> My assumptions are:
>  * I trust no single source of Random Numbers.
>  * I trust at least one source of all the sources.
>  * no particular difficulty with lossy combination.
I take the last item to mean that you do not mind wasting entropy but
want to be sure the resulting random number is unpredictable.

If you add one additional assumption:
* The sources are independent of each other
then the XOR of the random sources will be at least as unpredictable as
the most unpredictable individual random source (to keep away from the
entropy discussion). As far as I can se, this the "if at least one
source is unpredictable for a workload of x, the resulting random is
also at least that unpredictable" property that you seem to be looking

If the sources are not independent, in the most extreme case: the sources
are the same, the result is not so good. XORing in the same RNG stream
twice, however good the RNG, is not so useful ;-)

Without the threatmodel, I am not sure if this is a problem for you, but 
if the attacker has control or knowledge of some of the sources, he 
also knows the XOR of the remaining ones. In the case he knows all but
one sources, and the remaining source is not so unpredictable (LFSR,
poorly biased noise source), the result can be quite predictable (and in
weak RNG designs, the remaining source might be compromised).
Note that this could also be used to force the combined RNG to more
likely generate a chosen output.

Using hashfunctions to combine the randoms makes it computationally
harder for such chosen results to be generated, it quickly becomes
effectively a search problem for hash-collisions where you have only
limited choice on the input. Also temporary lulls in the quality of the
random sources are much better handled. Peter Gutmann's dissertation
has a very good description of what he did for hardening his cryptolib's
the random generation from many such attacks/mistakes.

With kind regards,
Wouter Slegers

The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to majordomo at

More information about the cryptography mailing list