Re: [Roll] Trickle clarification: (in)consistency

["Mathilde Durvy (mdurvy)" <mdurvy at cisco.com>]

Hi Pascal, Jonathan,

Thanks for your answers. I think we agree that child and other nodes
could play a role to detect a consistent situation. 
Hence, we need to see how this affects the DIO processing rules as you
actually typically don't process DIOs from node deeper than you in the
DAG...
I also agree with Pascal and Julien when they say that consistency
should concentrate on global or common DAG information.

Best,
Mathilde
-----Original Message-----
From: Pascal Thubert (pthubert) 
Sent: lundi, 23. novembre 2009 08:55
To: Julien Abeille (jabeille); 'Jonathan Hui'; Mathilde Durvy (mdurvy)
Cc: 'roll at ietf.org'
Subject: RE: [Roll] Trickle clarification: (in)consistency

Hi Julien

Seems to me that so far we are saying the same thing.

We'll be concentrating the DAG information in one option. That option
should be consistent in an iteration. It should thus be subject to
elision if it is redundant with C messages already heard during the
trickle period. Other fields in the DIO might not be redundant though.
My main concern is the routing information. My Rank is not my neighbor's
Rank.

Pascal

>-----Original Message-----
>From: Julien Abeille (jabeille)
>Sent: dimanche 22 novembre 2009 22:55
>To: Pascal Thubert (pthubert); Jonathan Hui; Mathilde Durvy (mdurvy)
>Cc: roll at ietf.org
>Subject: RE: [Roll] Trickle clarification: (in)consistency
>
>Hi Pascal,
>
>Not sure I understand your explanation. In a DIO there are informations

>that are global to the DAG and some that are node specific. We use 
>trickle because we consider DIO sending a consitency problem, hence we 
>focus on the global information. They come both from parents and
childs, so I would also increment C when hearing from a child. Am I
missing something?
>
>Julien
>
>> -----Original Message-----
>> From: roll-bounces at ietf.org [mailto:roll-bounces at ietf.org] On Behalf 
>> Of Pascal Thubert (pthubert)
>> Sent: vendredi 20 novembre 2009 20:12
>> To: Jonathan Hui; Mathilde Durvy (mdurvy)
>> Cc: roll at ietf.org
>> Subject: Re: [Roll] Trickle clarification: (in)consistency
>>
>> Hi Mathilde;
>>
>> I agree with Jonathan that we want to avoid redundancy.
>>
>> The rank of metric that a node advertises is an information about 
>> self that cannot be inferred from the neighbors so it is *not* 
>> redundant. If a number of nodes go down at the same time and one does

>> not advertise, then that node will look better and will attract the 
>> traffic. It will also attract the neighbors that went down and that 
>> will be tempted to reattach to that apparent survivor.
>>
>> The information in the DIO that relates to the DAG configuration will

>> be centralized in an option that IS redundant. We could add text that

>> a node that see C of it MAY filter it out from its DIO.
>>
>> Finally the new sequence is redundant. I proposed to use it to limit 
>> the density of routers using trickle. When you hear C router 
>> advertising a new sequence you are entitled to stop being a router 
>> for that sequence (stop trickle timer). The trickle timer would then 
>> be restarted upon a DIS to serve orphans, and upon a next sequence to

>> redistribute the load.
>>
>> Pascal
>>
>> >-----Original Message-----
>> >From: roll-bounces at ietf.org [mailto:roll-bounces at ietf.org]
>> On Behalf Of
>> Jonathan Hui
>> >Sent: vendredi 20 novembre 2009 16:41
>> >To: Mathilde Durvy (mdurvy)
>> >Cc: roll at ietf.org
>> >Subject: Re: [Roll] Trickle clarification: (in)consistency
>> >
>> >
>> >Hi Mathilde,
>> >
>> >On Nov 19, 2009, at 9:24 AM, Mathilde Durvy (mdurvy) wrote:
>> >
>> >>> - when do we increment C?
>> >>
>> >> The node MAY increment C when it hears another DIO.
>> >>
>> >> I think the spec says that a node MAY increment C each
>> time it hears
>> a
>> >> "consistent DIO for this DAG from a DAG parent". What fields are 
>> >> included in the consistency check? If only generic DAG information
>> are
>> >> checked you might want to look at DIO from non-DAG parents, no?
>> >
>> >You're right, this text in the draft is incorrect.  There's an 
>> >important distinction to make here.  We want to quickly send DIO 
>> >transmissions (reset the Trickle timer) whenever
>> *inconsistencies* are
>> >detected.  But you want to suppress DIO transmissions (increment C) 
>> >whenever *redundancies* are detected.  While you can detect 
>> >inconsistencies by listening to your DAG parent(s),
>> redundancies will
>> >probably come from your siblings/cousins.
>> >
>> >We are looking to update the text around incrementing C and
>> make what
>> >constitutes a redundancy more precise.
>> >
>> >--
>> >Jonathan Hui
>> >
>> >_______________________________________________
>> >Roll mailing list
>> >Roll at ietf.org
>> >https://www.ietf.org/mailman/listinfo/roll
>> _______________________________________________
>> Roll mailing list
>> Roll at ietf.org
>> https://www.ietf.org/mailman/listinfo/roll
>>