On Jun 18, 2009, at 4:41 AM, Rémi Villé wrote:

> Hi,
>
> I would like to discuss about how the path cost is calculated from  
> the cost of its arcs.
> Currently this path is accumulated with an addition. I think it's  
> bizarre because we try to create the best path in term of ratio (msg  
> received/msf sent). So the cost of a path should be proportional to  
> the chance of a packet to reach the sink through this path, i.e. 1/ 
> q(a,b)*q(b,c) for the path (a,b,c), and not 1/q(a,b) + 1/q(b,c). (q  
> = PRR(a,b)*PRR(b,a)).
>
> I tried to find incoherent cases with this accumulation and I found  
> this one :
> 3 motes : a, b and c (a is the sink).
> q(a,b) = 0.5
> q(b,c) = 0.2
>
> The cost of (a,b,c) should be 1/(0.5*0.2) = 10, but the effective  
> cost is 1/0.5 + 1/0.2 = 7. (q(a,b,c) = 0.5*0.2 = 0.1).
> If we have cost(a,c) = 8, then q(a,c) = 1/8 = 0.125.
> We have q(a,c) best then q(a,b,c) but cost(a,c) > cost(a,b,c), and c  
> choose a bad best parent/path (the path (a,b,c) instead of (a,c)).
>
> If I take ETX instead of EETX (10/q) it doesn't change this reasoning.
>
> I would like to know if I miss something, maybe there's a good reson  
> to use the addition instead of multiplication to accumulate the cost  
> path, I would like to understand...

No -- the cost assumes retransmissions. So the expectation is that  
you'll have 5 transmissions on b,c and 2 on a,b, for a total of 7.

10 assumes no retransmissions, which, as your math shows, is a bad  
policy, as it leads to multiplicative rather than additive costs.

Phil
_______________________________________________
Tinyos-help mailing list
[email protected]
https://www.millennium.berkeley.edu/cgi-bin/mailman/listinfo/tinyos-help

Reply via email to