Re: strengthening /dev/urandom

From: Mok-Kong Shen (mok-kong.shen_at_t-online.de)
Date: 08/30/04 

Date: Mon, 30 Aug 2004 10:39:04 +0200

Guy Macon wrote:
> Mok-Kong Shen <mok-kong.shen@t-online.de> says...
>
>>Guy Macon wrote:
>>
>>
>>>Bryan Olson <fakeaddress@nowhere.org> says...
>>>
>>>
>>>>Guy Macon's case is a counter-example that disproves your argument.
>>>>It need not do anything else.
>>>
>>>Other counterexamples:
>>>
>>>Assuming XOR combining...
>>>
>>>Input a is true random every odd read and deterministic every even read,
>>>Input b is true random every even read and deterministic every odd read.
>>>The output is true random.
>>>
>>>Input a is true random except during startup, Input b is true random
>>>unless it has been running for hours and gets very hot. The output
>>>is true random.
>>>
>>>Inputs abcdefghijklmnop are each true random 90% of the time, but
>>>whether they are true random or determistic at any moment is random
>>>and independent. In this case, the output is not true random, but
>>>the entropy has been concentrated to be far higher than the 90%
>>>randomness available at any one of the inputs.
>>
>>What if the timing is such that the 10% periods of both
>>stream coincide at some time during the operation of the
>>process? Do you have a mechanism to ensure that this couldn't
>>happen? If no, you are xor-ing in such period twos streams,
>>none of which has full entropy and the very argument that you
>>gave in a previous post fails.
>
>
> You are merly restating what I said: "In this case, the output
> is not true random." It has some non-random bits, but fewer
> than any of the inputs have. Thus it distills entropy, but not
> perfectly.
>
> Do you deny that every one of my examples distills entropy?
>
> (Distill entropy = make the percentage of entropy at the
> output higher than it was at any of the inputs, at a cost of
> not putting out as many bit as are coming in)

'Assuming' that the analogy of chemical distillation were
possible in our case, i.e. the software could distinguish
streams with entropy from ones without entropy, 'then' you
would be certainly right. See also another post, sent
simultaneously.

M. K. Shen

Loading