Re: [Roll] need clarification: DIS processing

[Tim Winter <wintert at acm.org>]

Hi Julien,

Julien Abeille (jabeille) wrote:
> of course it needs jitter. Now trickle is armed anyway at all times
> right (T timer). If we do not do anything specific, the DIO will not
> come earlier with DIS than without...
> So if no immediate response is needed, then DIS is not needed. I still
> think it is, for instance anytime you lose you r last parent, and do not
> want to wait possibly hours to get a DIO. This is the same as with RS/RA
> (jittered !)
>  

Yes, I agree DIS is needed also- and this is exactly how we got to DIS.  It
started by first allowing nodes to actively probe the neighborhood by using
`RS' to solicit `RA-DIO'. from their neighborhood in previous RPL versions.
 This was thought to be useful, e.g., for battery powered nodes that do not
want to wait.  Now that RPL does not use RS/RA/NA we have introduced DIS for
this exact purpose.

> Resetting trickle (I to Imin, and T to random ) does not make sense I
> guess, as sending a DIS does not always mean global instability.
> Thoughts on how to arm a small jittered timer?

It doesn't mean global instability- but it does mean that something has
changed in the local neighborhood at least.  So fair enough I think to reset
trickle, to let trickle do it's job.  I agree with prior comments- lets not
make a special case here for DIS handling.

The DIS should then also be multicast, and it should trigger multicast DIO
responses.

Regards,

-Tim


>  
> Best,
> Julien
> 
>     ------------------------------------------------------------------------
>     *From:* Pascal Thubert (pthubert)
>     *Sent:* lundi 9 novembre 2009 13:53
>     *To:* Julien Abeille (jabeille); Philip Levis
>     *Cc:* ROLL WG
>     *Subject:* RE: [Roll] need clarification: DIS processing
> 
>     At the moment, I cannot figure when an immediate response would be
>     needed.
> 
>      
> 
>     I think Phil mentions that somewhere else as well. The idea would be
>     to arm trickle and let it time out.
> 
>      
> 
>     Also let us keep in mind that in networking at large, most timers
>     are jittered to avoid network synchronizations. In wireless, it is
>     even worse since as you point out, synchronization also means
>     collisions.
> 
>      
> 
>     Pascal
> 
>     *From:* roll-bounces at ietf.org [mailto:roll-bounces at ietf.org] *On
>     Behalf Of *Julien Abeille (jabeille)
>     *Sent:* lundi 2 novembre 2009 18:01
>     *To:* ROLL WG
>     *Subject:* [Roll] need clarification: DIS processing
> 
>      
> 
>     Hi all,
> 
>      
> 
>     one question:
> 
>     when a router receives a DIS, should it answer right away? I guess
>     this is a good source of collision, so a random timer would help.
>     Could we set T to a random value between I_min /2 and I_Min, without
>     resetting I (receiving a DIS does not mean inconsistency)?
> 
>      
> 
>     Best,
> 
>     Julien
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Roll mailing list
> Roll at ietf.org
> https://www.ietf.org/mailman/listinfo/roll