[Olsr-dev] About RSSI using / better minstrel

[Teco Boot]

Op 4 jan. 2012, om 09:31 heeft Bastian Bittorf het volgende geschreven:

>> Such as two paramenter will be configured: rssi_low, rssi_high.
>> 
>> If rssi on one link is more than rssi_high, then this link is active; if
>> rssi on this link is less than rssi_low, 
> 
> IMHO this is not a good idea:
> 
> 1) a good RSSI does'nt mean that we have good speed
> 2) a good RSSI does'nt mean, that we find a network-path,
> that fits our needs (speaking: the destination route what we
> are searching for, e.g. 0.0.0.0/0 or a specific HNAx)
But it has also advantages:
1) RSSI doesn't depend on user traffic
2) it reacts quickly on link changes

> Better make a metric-plugin, which works like etx_ffeth,
> but take the speed-values from minstrel[1] into account.
Yes, lets do this.
We can use a composed metric, based on:
A: LQ / NLQ  (adjusted ETX, but *not* symmetric)
B: Minstrel BestThroughput (Tx side)
C: RSSI (Rx side)
Question is: how to bring stats to neighbor?

The easiest calculation is something like: (A + B + C ) / 3
More advanced is a table lookup, with values based on A and B 
and using C as pointer. And auto-calibrate the table.

Dampening and hysteresis is crucial (of course).

It would be nice (required?) to keep interoperability with neighbors
not having RSSI and / or Minstrel, or even with etx_ffeth.

> problem here: minstrel only works, if there is traffic, so a
> yet unused route defaults to e.g. lowest basic-rate (1/6mbit).
> a simple approach would be to generate traffic for each neighbour
> when they appear initially and on regular intervals.
Yepp.
I run a script with background probing. Watch out for alternative paths, e.g. 
Ethernet or 2-hop path to neighbors

> problem2: minstrel only exist on linux, so the others are out
> (but they are out anyway... 8-)
Is BestThroughput made available in Linux WLAN stack? I used to capture 
MadWifi statistics, but with iw I lost this. Recompile kernel with debugfs 
is not my favorite.

Teco 

> 
> bye, Bastian
> 
> 
> -- 
> Olsr-dev mailing list
> (spam-protected)
> https://lists.olsr.org/mailman/listinfo/olsr-dev