Oliver Xymoron wrote:
I'm not an expert in this field, so bear with me if I make any blunders
obvious to one trained in information theory.
> The current Linux PRNG is playing fast and loose here, adding entropy
> based on the resolution of the TSC, while the physical turbulence
> processes that actually produce entropy are happening at a scale of
> seconds. On a GHz processor, if it takes 4 microseconds to return a
> disk result from on-disk cache, /dev/random will get a 12-bit credit.
In the paper the accuracy of measurement is 1ms. Current hardware has
tsc precision of nanoseconds, or about 6 orders of magnitude more
accuracy. Doesn't this mean that we can pump in many more bits into the
algorithm and get out many more than the 100bits/min that the setup in
the paper acheives?
> My entropy patches had each entropy source (not just disks) allocate
> its own state object, and declare its timing "granularity".
Is it that straightforward? In the paper they go through a number of
stages designed to remove correlated data. From what I remember of the
linux prng disk-related stuff it is not that sophisticated.
> There's a further problem with disk timing samples that make them less
> than useful in typical headless server use (ie where it matters): the
> server does its best to reveal disk latency to clients, easily
> measurable within the auto-correlating domain of disk turbulence.
If this is truly a concern, what about having a separate disk used for
nothing but generating randomness?
Chris
-- Chris Friesen | MailStop: 043/33/F10 Nortel Networks | work: (613) 765-0557 3500 Carling Avenue | fax: (613) 765-2986 Nepean, ON K2H 8E9 Canada | email: cfriesen@nortelnetworks.com- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Oct 31 2002 - 22:00:43 EST