Re: [Roll] knowing which multiple metrics matter: MRHOF related questions

[Ralph Droms <[e-mail]>]

On May 28, 2012, at 3:04 PM 5/28/12, Pascal Thubert (pthubert) wrote:

> Hello Micheal
> 
>>>>>> "Pascal" == Pascal Thubert <(pthubert)" <[e-mail]>> writes:
>    Pascal> I think RPL does not want to take party there. The OF is a
>    Pascal> piece of logic to tie metrics and policies together.  
> 
> My question is:
> 
> - do the nodes of a DODAG have to use the same metrics to pick a parent,
>  (and if so, how do they agree)
>  I think that they do not, so long as they use an algorithm (such as
>  MRHOF) which has certain properties.
> 
> [Pascal]  The current spec requires that all players of an instance play by the same rule, that is same OF, in order to guarantee the expected result. 

Well ... given the current definitions of OF, I would say that even specifying the same OF across a RPL instance does not guarantee a consistent result.  OF0 allows pretty much any mechanism at all - within some constraints on the range of the computed rank - so there are no guarantees about how different nodes will compute their ranks under OF0.  If different nodes use a different metric under MRHOF, there will be no guarantees of consistency for MRHOF.

> The goal there was deeper than the OF, like same OF with same parameterization. MRHOF is generic and allows various incarnations, depending in the metrics; all those incarnations are to be seen as different OFs WRT to RPL 's uniqueness rule.

So, MRHOF+ETX is a different OF than MRHOF+link-quality?

> Note that this rule is not for looplessness, the Rank would assure that anyway.

OK.

> - if we had multiple RPL instances in an LLN, using different metrics,
>  then we would have multiple RPL Instances and DODAGs.  The different
>  set of metrics would not co-exist in the same RPL Instance.
> 
> [Pascal] I think we agree. An instance can use multiple metrics bound by some logic. But those metrics and that logic must be identical throughout the instance. 
> A different (set of) metric means a different OF even if MRHOF is behind both.

I'm not sure how to reconcile your first two sentences with the MRHOF spec.  As I read the MRHOF spec, any individual node uses a single metric to compute its rank.  I infer that all the nodes in a RPL instance (using MRHOF) would use the same metric.

I understand that other OFs might use a rank computation that combines several metrics.  Do I have it right that MRHOF uses a single metric?

- Ralph

> 
> Cheers,
> 
> Pascal
> 
> -- 
> Michael Richardson <[e-mail]>, Sandelman Software Works 
> IETF ROLL WG co-chair.    http://datatracker.ietf.org/wg/roll/charter/