sta_info->fail_avg

sta_info->fail_avg

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 219 bytes --]

Hi Luis,

Can you explain what you're using this for in the mesh code? I can't
figure it out, but I do know that neither ath9k's nor Intel's rate
control algorithms ever set it so it'll always be zero.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: sta_info->fail_avg

[Javier Cardona]

Johannes,

On Wed, Sep 10, 2008 at 4:59 PM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> Hi Luis,
>
> Can you explain what you're using this for in the mesh code?

It is used to estimate the average retransmissions that will be
required to send a nominal frame over a given peer link.
That's one component of the airtime link metric.

> I do know that neither ath9k's nor Intel's rate
> control algorithms ever set it so it'll always be zero.

In that case transmission failures will not be taken into account and
nodes with that hardware will report a lower (better) link metric than
what they should.  In other words, those nodes will attract mesh
traffic, which will provide a strong incentive for the driver
maintainers to set that value correctly :)

Cheers,

Javier
-- 
Javier Cardona
cozybit Inc.

Re: sta_info->fail_avg

[Luis R. Rodriguez]

On Wed, Sep 10, 2008 at 10:15:12PM -0700, Javier Cardona wrote:
> Johannes,
> 
> On Wed, Sep 10, 2008 at 4:59 PM, Johannes Berg
> <johannes@sipsolutions.net> wrote:
> > Hi Luis,
> >
> > Can you explain what you're using this for in the mesh code?
> 
> It is used to estimate the average retransmissions that will be
> required to send a nominal frame over a given peer link.
> That's one component of the airtime link metric.
> 
> > I do know that neither ath9k's nor Intel's rate
> > control algorithms ever set it so it'll always be zero.
> 
> In that case transmission failures will not be taken into account and
> nodes with that hardware will report a lower (better) link metric than
> what they should.  In other words, those nodes will attract mesh
> traffic, which will provide a strong incentive for the driver
> maintainers to set that value correctly :)

Neither of our drivers support Mesh yet though so this is good to know,
once we do add it (or someone does).

  Luis

Re: sta_info->fail_avg

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 1146 bytes --]

Javier,

> It is used to estimate the average retransmissions that will be
> required to send a nominal frame over a given peer link.
> That's one component of the airtime link metric.

Ok.

> > I do know that neither ath9k's nor Intel's rate
> > control algorithms ever set it so it'll always be zero.
> 
> In that case transmission failures will not be taken into account and
> nodes with that hardware will report a lower (better) link metric than
> what they should.  In other words, those nodes will attract mesh
> traffic, which will provide a strong incentive for the driver
> maintainers to set that value correctly :)

Heh, ok. I've been trying to reduce the amount of stuff the rate control
algorithm needs to do, does this need to be in the rate control
algorithm? It seems that the we simply calculate

pf = spinfo->tx_num_failed * 100 / spinfo->tx_num_xmit;

and

fail_avg = ((pf + (spinfo->last_pf << 3)) / 9);
(in a fixed-point format that has more precision)

and we could do all that in the generic code if we get proper
information from the driver. Anyway, for now I'll leave it as-is.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]