Re: [Roll] Clarification on the objective function

["Julien Abeille (jabeille)" <jabeille at cisco.com>]

Hi Phil,

> -----Original Message-----
> From: Philip Levis [mailto:pal at cs.stanford.edu] 
> Sent: lundi 9 novembre 2009 19:58
> To: Julien Abeille (jabeille)
> Cc: Pascal Thubert (pthubert); JP Vasseur (jvasseur); 
> Mathilde Durvy (mdurvy); roll
> Subject: Re: [Roll] Clarification on the objective function
> 
> 
> On Nov 9, 2009, at 10:32 AM, Julien Abeille (jabeille) wrote:
> 
> > Phil,
> >
> > Are you in favor of computing rank based on all parents, or only 
> > preferred?
> 
> It's not clear to me what a "preferred" parent is. Is it the 
For me the preferred parent is the one you send DAOs to. 

> exact parent at time T? A node may wish to change which next 
> hop it uses very quickly (albeit under constraints that keep 
> the topology consistent). Jonathan Hui's IP layer does this: 
> it chooses an alternate parent after three consecutive link 
> delivery failures (no acks).
agree
> 
> The way I see it, a node has a neighbor set, which is an 
> unconstrained set of link-local neighboring nodes. Its parent 
> set is a subset of its neighbor set, and these parents can be 
> used arbitrarily (an implementation could just cycle through 
> them for each transmission, if it wanted). 
agree

> The only time when 
> a node has *a* parent is the instant it transmits a packet to 
> a particular next hop.
> 

> Given this, Rank should be computed based on the parent set. 
> I'd say that
> 
> A node MUST NOT advertise a Rank that is less than or equal 
> to any member of its parent set.
> 
Agree. 
The issue is when you lose your prefererred parent, you may want to
choose a backup one to send DAOs to. The fact that its rank is lower
than yours does not guarantee that your new rank will not change. e.g.
if OF says my rank is preffered parent's + 2, and I have parents at rank
1 and 2, mine is 3. if my parent at rank 1 dies, and I choose the other
parent as preferred, my rank would become 4, which implies moving lower.
Does anybody want to avoid this by computing continuesly rank based on
the parent set instead of just the preferred one (this is not optimal
but helps if you compute your rank when you get several DIOs)? 

Julien
> Phil
>