faster /dev/random

Wolfgang S. Rupprecht wolfgang.rupprecht at gmail.com
Sun Aug 22 07:54:29 UTC 2010


Tim <ignored_mailbox at yahoo.com.au> writes:
> On Sat, 2010-08-21 at 23:43 -0500, Thomas Cameron wrote:
>> I've always heard that you can get faster random numbers by generating
>> a lot of interrupts.  I usually do something like
>> run /etc/cron.daily/mlocate to generate a lot of disk activity.
>
> I've always wondered whether that would provide it with real random
> seeding, or seeding with a pattern that might be determinable.  After
> all, you're basing it on the disk contents.

Now that you mention it, I wonder how that works too.

The inter-interupt timing for the disk will probably also be quite
biased.  Interrupts that come at a one per sector rate will probably
have strong spikes in their probability functions for "(rotation-time /
number-of-sectors-per-track)".  Even with disk zones, there probably
won't be that many different sectors-per-track values.

-wolfgang
-- 
Wolfgang S. Rupprecht      http://www.wsrcc.com/wolfgang/      (IPv6-only)


More information about the users mailing list