Sorry to be beating you up today on
the mailing list. It's not intentional; maybe it's just your turn
:)
In Geneva, I thought we had reached
an epiphany when Thomas noted so adroitly that destination nodes cannot
a priori know what source nodes might be sending them packets in the future.
I believe he referenced the pain Google would have if it somehow
needed to define a connection to potentially everyone wanting to someday
access its website!!! I hoped that discussion would be the impetus
to find an on-demand means to pass a packet outward. The idea of
every node having to form a connection to every other node in the DAG to
be maintained ad infinitum on the off-chance that someday a packet may
need to be passed to that node is very inefficient.
As probably the strongest proponent
of the P2P requirement on this mailing list, I will readily admit that
even in building control only about 25% of the router nodes need to pass
intra-LLN P2P messages. As I understand the DAO algorithm in RPL
4.0 using the preferred parent requirement there will be one and only one
outward path defined to each node. If that lonely path should happen
to be dropped, the outbound packet would be in a limbo state for an undetermined
amount of time until the DAG recognizes the problem, issues a sequence
number change and the DAG is fully repaired with a new set of DAO exchanges.
The source node will have long ago exhausted its retry count and
flagged the destination node off-line.
So yes, I think we either need to design
outward traffic into the DAG at the same level as we do inward traffic;
or devise an on-demand scheme where a node can easily 'discover' another
node in the DAG as needed. Richard's comment below is profound saying
that right now establishing a good outward link is a matter of luck, not
design.
Regards,
Jerry
JP Vasseur <jvasseur at cisco.com> Sent by: roll-bounces at ietf.org
11/19/2009 08:49 AM
To
Richard Kelsey <richard.kelsey at ember.com>
cc
roll at ietf.org
Subject
Re: [Roll] updating DAO caches (was
Re: Something to ADD)
On Nov 19, 2009, at 3:41 PM, Richard Kelsey wrote:
>> From: JP Vasseur <jvasseur at cisco.com>
>> Date: Thu, 19 Nov 2009 13:18:56 +0100
>>
>> On Nov 19, 2009, at 12:01 PM, Richard Kelsey wrote:
>>
>> JP> From: JP Vasseur <jvasseur at cisco.com>
>> JP> Date: Thu, 19 Nov 2009 11:36:49 +0100
>> JP>
>> JP> I do see several deployment cases where no storing DAO
would
>> lead to
>> JP> extremely sub-optimal paths and even more importantly traffic
>> JP> congestion when getting closer to the root of course.
>>
>> Richard> Can you share these cases with us? I agree with
Jonathan
>> Richard> that storing DAO states will typically not provide
much
>> Richard> improvement in P2P routing.
>>
>> Sure. I'll take the example of an inter primary+secondary
>> substation network (could be the smart metering network)
>> that supports traffic for various purposes: meter
>> read-out, DA alarms, ... etc. Thus traffic of various
>> nature: P2MP, MP2P and P2P.
>
> Yes, I think we all agree that reasonably efficient
> P2P is a requirement for many use cases.
>
> The issue isn't whether or not we need good P2P routing.
> The issue is whether or not the DAO mechanism does a good
> enough job to be worth the effort.
>
> RPL ignores P2P when choosing parents. Getting a good P2P
> route out of RPL is a matter of luck, not design.
>
First, you can certainly tune the OF to increase the P2P quality.
Second, in the VERY worst non probable case you would transit via
the route.
Last but not least, without DAO how do you send traffic outward ?
Need to wait until you receive a packet, then record route and do
source routing ?
>> There are many such networks
>> where not storing DAO just does not work since all P2P
>> traffic will have to transit via the root (unacceptable
>> delays for alarms) and the traffic around the root will be
>> way too high.
>
> Yes, and in many cases most P2P traffic will have to transit
> via the root even if all nodes store DAOs. This is a side
> effect of how RPL chooses parents. Minimizing the lengths
> of the paths to the root maximizes the number of pairwise
> paths that include the root. If this is unacceptable, then
> we should stop tweaking the DAOs and work on a better
> solution.
>
-Richard Kelsey
> _______________________________________________
> 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
