RE: WG ACTION: 2 weeks to discuss [LL62] Do not space probes randomly

> Christian, we can answer this very simply. I gave a little picture
> showing the probability density function for the simple scheme as we
had
> it two years ago:
> 
> 
>     p|
>      |
>      |
>      |    ----------------
>      |    |              |
>      |    |              |
>      |    |              |
>      |----|----|----|----|----|----|
>     -1    0    1    2    3    4    time (seconds)

If I understand correctly, the scheme that you describe here assumes an
initial spacing over 1 second, followed by two repetitions spaced one
second apart. You measure the probability that a packet be sent on any
given second. Since in your scheme each second is a perfect repeat of
the previous one, the scheme is indeed flat. However, having each second
be a perfect repeat of the previous one is not a very desirable
characteristic.

The scheme that I described would indeed have a different curve -- just
as wide, but taller. It will however minimize the probability of
collisions.

If you do want a scheme that is also flat but random, then you can do
the following:

	Wait R1 = random(0..1) before the 1st try;
	Wait R2 = 1+random(0..1)-R1 before the second try;
	Wait R3 = 2+random(0..1)-R2 before the third try.

It produces the same flat distribution that you find desirable, without
the correlation between the consecutive intervals.

-- Christian Huitema