[Olsr-dev] Seeking comments: OLSR+ETX v/s DSR+ETX

Sven-Ola Tuecke (spam-protected)
Wed Dec 19 13:17:02 CET 2007


my last post has some serious math flaws as well as a case of mistaken OP 
identity. Sorry for that (especially to seb) - heres the corrected version.

<Reworked Post>
I don't know, what parts of the current olsrd implementation you denote with
the term "voodoo". Link quality is a number between 0.0 and 1.0 and it
reflecs how many packages are lost between sender and receiver. How LQ is
evaluated/measured may #include <voodoo.h> (especially the 128er bit mask
array) but nothing that really deserves that term IMO.

The rest is a formula, commonly known as ETX. Which is optimized / meant for
meshing with single interfaces. As we all know, ETX = 1 / (LQ * NLQ). And 
they're summed up to form a value for a rouing path with 
PathETX=ETX1+ETX2+...+ETXn. The plus sign in the latter calculation has a 
magic '1' included for every hop, so it favors routes with less hops...

This calculation overcomes a small weakness in LQ-only measurements: If a 
amount of links in a path do not have packet loss, the second-level criteria
(Hops) works well for route decisions. Moreover - there where talks about
exchaning that plus with a multiplication operator (hence something like
PathEtx=ETX1*ETX2*...*ETXn) or introducing some "voodoo factors" to have a
seamless/tweakable transition between '+' and '*'. But I think that's
another story...
</Reworked Post>

// Sven-Ola

More information about the Olsr-dev mailing list