[Battlemesh] battlemesh v6 test data postprocessing

[Gui Iribarren]

On 07/06/14 20:10, Juliusz Chroboczek wrote:
>> Yes, for netperf_test0 and netperf_test1 it was set to 24Mbps, at the
>> end of netperf_test2 the multicast rate was set from 24 to 54Mbps, on
>> netperf_test3 it was set to 6Mbps.
> 
> Ah, okay.  That probably explains it.
> 
> Babel assumes that multicast packets get through often enough (at least
> one in three), even if they're full-size frames (1500 octets).  That's
> probably not going to happen with such extreme parameters.
> 
> Clauz, are these parameters typical of community mesh networks?  If so,
> I'll add some code to Babel to deal with multicast rates set too high --
> but I'd obviously rather not add code just to deal with Battlemesh.

Yes, in deltalibre we use 36mbps for 2.4ghz and 24mbps for 5ghz, in most
cases

nodes are set 100 - 300mts apart, in long lines
with mcast_rate=1mbps, batman-adv greatly overestimates the link quality
between faraway nodes, and ends up using long links that have poor
throughput

quintanalibre, a small town in a mountain environment, uses 24mbps /
12mbps respectively AFAIR

in both networks, all (or most) nodes are dualband routers (tl-wdr3500)
so there's minimal bandwidth degradation on each hop, so several-,
short-hop paths have much better perfomance than few-, long-hop paths.

a medium-sized qMp mesh (50 nodes) in barcelona, uses bmx6 (layer3) but
also faced similar issues with mcast_rate=1mbps, they have raised it a
year ago but i'm not sure to which value.

cheers!

> 
> -- Juliusz
> _______________________________________________
> Battlemesh mailing list
> Battlemesh at ml.ninux.org
> http://ml.ninux.org/mailman/listinfo/battlemesh
>