Re: Random device in 1.3.31 on [34]86...

Marek Michalkiewicz (
Fri, 13 Oct 1995 14:25:04 +0100 (MET)

Theodore Ts'o:
> The problem is that the timer is very slow to access, so it
> significantly increases your interrupt latency; unfortunately, this
> happens on the slower machines which may ill-able to afford the extra
> speed hit. Without the timer, randomness accumulates much more slowly,
> but it doesn't hurt your system as much.

How about this:

- only use the timer if we really need more randomness (entropy_count is
below some threshold value) - no extra overhead if no one is using the
random device at the moment

- move the add_interrupt_randomness call after the interrupt handler proper
(the same overhead but less interrupt latency)

That seems like a reasonable compromise to me, still cheaper than to buy
a Pentium :-).