[Battlemesh] Another peer-reviewed assessment of Babel, Batman and OLSR

[Juliusz Chroboczek]

Dear all,

I've just stumbled across another peer-reviewed experimental evaluation
of our favourite routing protocols.  I haven't found an authoritative
bibliographic entry (do the authors perhaps not want to be cited?), so
the following should be taken with a pinch of salt:

  David Murray, Michael Dixon and Terry Koziniec.  An Experimental
  Comparison of Routing Protocols in Multi Hop Ad Hoc Networks.  In
  Proc. ATNAC 2010.  2010.

You'll find a copy online on

  http://bridgingthelayers.org/docs/murray_routing.pdf

I rather liked this paper.  The introduction indicates that the authors
actually know what they're speaking about -- I've only found two claims
that I didn't agree with (that CPU time is a limiting factor on modern
embedded systems, and that BATMAN is ``a new and different approach'' --
I think of it as a variant on the distance-vector theme).  They actually
tested OLSR-ETX, which is the right choice according to Aaron et al,
although they call it OLSR in the paper.  There's also an interesting
discussion of the L2 vs. L3 issue.

The meat is on pages 5 and 6.  A few choice quotes:

  * ``Results varied for the different topologies however all packet
    delivery ratios were between 99.6% and 99.98%.''  (This confirms our
    results from WBMv2 -- all of our routing protocols work.)

  * ``The results of the bandwidth tests, shown in Fig 3, reveal that
    the Babel routing protocol provided better throughputs than OLSR,
    BATMAN L3 or BATMAN L2.''  (This is surprising -- I'd expect everyone
    to have roughly the same bandwidth -- and I think it might be an
    artefact of the way Babel measures packet loss, using piggy-backed
    Hellos.)

  * ``OLSR transfers the greatest number of bytes of routing protocol
    overhead.  Comparatively, Babel produces a minuscule overhead. [...]
    It is evident that BATMAN transmits the largest number of routing
    packets/frames.''  (Since they used rather good, stable links,
    Babel's route request mechanism never triggered.  This is unlike
    WBMv2, where the horribly bad routes we were measuring with caused
    nodes to oscillate in and out of range, which is Babel's worst case.)

                                        Juliusz