Re: [Roll] I-D Action: draft-ietf-roll-p2p-rpl-10 comments

[Mukul Goyal <[e-mail]>]

> [Federico]
>> The situation that I describedcan be seen from another point of view.
> Suppose now that A and B got about 15-20 neighbor.
> A got a rank of 500
> B got a rank of 3000
>
> A suppresses DIOs for the reason described previously so node B will
> continue to use a bad path for a very long time because it does "not
> know" that node A "exists". I think that K should be at least =2 or more.
>
> [Mukul]
> Continue using bad path? The route discovery is still going on! In the scenario you described, B probably will take a long time to be "discovered" or perhaps not be discovered at all within the specified DAG lifetime.

[Federico2]
Exactly, do not you think that the fact that node B take a long time to 
discover node Ais a problem?

[Mukul2]
No, it is not if B also has 15-20 neighbors like A. As long as the origin discovers a path to B that meets the specified constraints, there is no problem. It does not matter if the discovered path is through A or not.

[Federico2]
 Suppose that node B wants to communicate 
with a neighbor of node A, it could lead to sending many unnecessary DIOs.

[Mukul2]
You described two separate scenarios:
1) A has 15 neighbors, B has just A as the neighbor.
2) Both A and B have 15 neighbors including each other.

In the second scenario, K=1 definitely makes sense because there are many candidate paths and one such path to B would probably be discovered even though many of B's neighbors would end up suppressing their DIOs. It is possible that the discovered route wont pass through A.

In the first scenario, discovery of a route to B depends on A making a DIO transmission and B receiving it. This would be delayed if A ends up suppressing its DIO again and again. The route discovery would fail if DAG lifetime is over before B could receive the A's DIO. If that is very likely to happen in a particular deployment, K should not be 1. No doubt about it. But, we cannot set the K value in default configuration option to account for this case, which is certainly not the common case.

Thanks
Mukul


>> On the other hand, if Imin/K are set less conservatively, you might see too many DIOs generated. There is no silver bullet. No single setting would work well in all situations. We do provide some guidance on how to set trickle timer in Section 9.2 of the draft. The trickle parameter values in default configuration option are a conservative estimate based on the desire to avoid too many DIOs. If these settings are not OK for a particular deployment, well then the default configuration option should not be used. The connectivity variance you described is indeed a tricky situation and would require a tradeoff with the need to avoid generating too many DIOs. I imagine one possible solution could be to allow a node to set its k based on its position (rank) in the DAG (k increases as you move closer to the edge). But, again there might be cases where this (or any other fix) wont work.
>>
>> Thanks
>> Mukul
>>
>> ----- Original Message -----
>> From: "Federico Consoli"<[e-mail]>
>> To: [e-mail]
>> Sent: Saturday, May 5, 2012 4:04:09 AM
>> Subject: [Roll]  I-D Action: draft-ietf-roll-p2p-rpl-10 comments
>>
>> Hi,
>> I disagree with the default value of trickle timer, especially for
>> Imin=6 and K=1.
>> If you got a dense network it could lead to a very long time to build DAG.
>>
>> Eg. Suppose you got 2 node A, B.
>> Node A got about 15-20 neighbor
>> Node B got only A as a neighbor.
>>
>> When node A receive its first DIO it resets trickle to Imin and it
>> starts trickle timer. Since A got a lot of neighbor its probably receive
>> another consistent DIO so it will suppress the DIO message and it will
>> increase its Imin to 7. So if node A suppress 6-8 times its DIO node B
>> will receive it after a long time. Finally if the link A-B got an ETX of
>> 2 or 3 (or more) this time doubles.
>>
>> These situations are not uncommon, especially in the routers at the edge
>> of dense networks.
>>
>


-- 
Regards
Consoli Federico