Re: [B.A.T.M.A.N.] Originators with cero TQ

[Antonio Quartulli <antonio@meshcoding.com>]

[-- Attachment #1: Type: text/plain, Size: 3253 bytes --]

On Mon, Oct 21, 2013 at 06:37:04PM +0200, Fernando Pizarro wrote:
> El 21/10/13 15:17, Antonio Quartulli escribió:
> > On Mon, Oct 21, 2013 at 01:23:37PM +0100, Bruno Alexandre Taraio dos Santos Antunes wrote:
> >> I think its spanish for zero
> > In that case..If I remember correctly you can have TQ=0 only for "Potential
> > Netxthops". If an originator has a TQ=0 on the best path then it is should not
> > be shown in the originator table.
> >
> > However, if you have further doubts, please post the table.
> >
> > Regards,
> >
> Hi all,
> 
> That's correct. Nodes with cero value (TQ=0) there are in "Potential 
> Nexthops" but there are a lot of them in the originator table. In some 
> cases, only there is one node with TQ major than 0. Example:
> 
> 172.16.0.9    0.300s   (196)       172.16.0.36 [     wlan1]: 172.16.0.40 
> (  0)       172.16.0.35 (  0)       172.16.0.22 ( 0)       172.16.0.39 
> (  0)       172.16.0.33 (  0)       172.16.0.32 (  0)       172.16.0.36 
> (196)
>

It depends on your network. There is no rule saying "there must be more than one
potential nexthops having TQ > 0". For sure you must have at least one with TQ
non zero (which would be your best nexthop).

> There are another cases, a behaviour that I don't understand is this: 
> Traffic is being routed using a SINGLE jump (or direct link) very low 
> bandwidth to its target, but there are many other options with better 
> bandwidth, however using more jumps/links, why is this happening and how 
> can I control it?

Bandwidth itself is not considered when computing the TQ.
The TQ is computed starting from a packet loss counter.
Moreover the protocol introduces a "hop penalty" which is used to avoid loops
and to push the protocol to use the shorter route in case of many "optimally
equal" paths.

> 
> Number of hop are more important than link quality to calculate TQ 

You can try to tune the value of the "hop penalty" I was talking about by using
the knob at "/sys/class/net/<mesh_iface>/mesh/hop_penalty". Its meaning is
described in Documentation/ABI/testing/sysfs-class-net-mesh.

This knob decides how much the TQ has to be reduced at each hop (remember that
there are other penalties too...). You can read this page to get some more
details about the hop penalty and more:
http://www.open-mesh.org/projects/batman-adv/wiki/OGM

However be sure that you know what you are doing when changing the hop penalty
value.

> value? Is there any "well-known formula" to calculate TQ?
> 

No "well known" formula, sorry. The TQ is the result of a number of operations
that you can read from the code and use to write your formula. So writing a
formula is still possible but it is not present in any documentation for the
reason I stated before.


> I make some tests changing default class of a client to 3 (batctl gw 
> client 3) with same results. What is the best client value for a large 
> enviroment (about 40 AP's)?

the class you are talking about is part of the gateway feature. Be sure you have
read this http://www.open-mesh.org/projects/batman-adv/wiki/Gateways before
going on. This feature does not affect the routing at all.

Regards,

-- 
Antonio Quartulli

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]