Minutes IETF98: roll
Routing Over Low power and Lossy networks
||Minutes IETF98: roll
ROLL WG Meeting
IETF 98 - Chicago
Thursday, 30 March, 2017, 17:40 - 18:40
Chairs: Ines Robles and Peter van der Stok
State of the WG 17:40 - 17:50 (10 min.) Ines & Peter
DAO projection 17:50 - 18:00 (10 min.) Pascal
Npdao optimization 18:00 - 18:14 (14 min.) Rahul
Load-balancing 18:14 - 18:24 (10 min.) Mamoun
AODV-RPL 18:24 - 18:33 (9 min.) Charlie
RPL-info 18:33 - 18:38 (5 min.) Michael
Q&A 18:38 - 18:40 (2 min.) Open Mic
[17:41] State of the WG 17:40 - 17:50 (10 min.) Ines & Peter
* Chartered deliverables: some delays.
* Lists WG drafts, related drafts, open tickets
[17:44] DAO projection 17:50 - 18:00 (10 min.) Pascal
* now a WG document
* recent addition of Source Routed projected state. Many new examples.
* main text focusses on the mechanism, example are pushed to Apendices.
* question to the audience: is there value to be able to project any type
DAO into the DODAG? * if multiple addresses in one VIO, means projected
path. * if multiple VIAs, means * question is how to compress all of this.
* Peter: like the projected DAO, but how does the root know how much space
is available in the node? * Pascal: so far considered out of scope,
external means such as Management. However, if situation is dynamic or if
multiple roots can inject state, we need a protocol to learn about
remaining space. If people interested in thi scapability let it be known
on the mailing list.
* IP-IP encapsulation
* e.g. a 16-hop route, want to install for route segments of 4 hops.
One address replaces 4. This is different from encapsulation
* discussion had on the mailing list: how is topology known to the root?
node capabilities known to the root? do we want mixed mode? * comment on
the mailing list about these topics. * Pascal goes through example.
Compares path stretch (node 51 to 53) in non-storing mode, storing mode,
and with new functionality in storing mode.
In non-storing mode, 41 will encapsulate and source-route to 43 via 42.
See the final slides online.
* Rahul: complexity of mixed mode when DAG has inconsistencies. E.g., a
loop. * Pascal: the new DAO would inject route of higher administrative
preference. * Rahul: no way to solve inconsistency. * Pascal: to be
discussed on the mailing list
[18:02] NP-DAO optimization 18:00 - 18:14 (14 min.) Rahul
* draft first presented in Buenos Aires. Applies to storing mode.
* simple solution to handle the problem
* a lot of care has been given in RPL to create routes, not so much on
removing them. This works addresses this. * currently, NoPath-DAO is sent
along the bad/failing/disapperaring path. This is going to fail by design.
Also ,NP-DAO and new DAO follow different paths, different delay, so
there may be a period of time during which there is no route at all.
* proposed change: the common parent is responsible for sending a NP-DAO
*down* the old path to prune the stale state. * Pascal: clarifying
question. This is pruning the broken branch from the trunk. Rahul cofirms *
This also allows invalidating routes for dependant nodes, not just
reporting node * Mcr: does this mean .. ? Rahul: D sends to A which send
NP-DAO down * MCR goes through example. What i traffic from J to C? ... *
Rahul: this improves the situation vs. previous situation. * This limits
the route downtime, but there * Pascal: moving between DODAGs, DODAGthe
roots exchange ND with new EARO option so the old root is aware that the
node has now attached to the new root. The expectation was that a flow
exactmy like this would clean up the stale routes in the old , same design
was done. So this is what we need. * Rahul: we'll also look at the Backbone
router. * changes: I flag to mean "invalidate on my behalf". R flag to tell
Reverse direction (i.e. going down).
[18:15] Load-balancing 18:14 - 18:24 (10 min.) Mamoun
* idea presented at a IEEE conference in oct 2016, paper on line
* RPL missing a mechanism to balance number of children.
* example presented, all children select a same parent, leads to dissymmetry
* Pascal: in storing mode, node 3 can run out of memory. There is (could
be) an overload bit in DIO to tell "attach to another node".
Regrading the saturatino of throughput, use the throughput metric.
* Dominique: to solve the problem of ndoes running out of energy, use the
remainning energy metric. * Rahul: allow for nodes to count the number of
children without creating new messages or having extra traffic. * propose
to have DIO include an option with IP address of parent. Parent will
process DIO even though rank is greater and recognize its address. * IOW
this leverages a DIO that would be ignored otherwise. * Pascal: you're
showing trees. RPL is about DODAGs. Nodes in the middle would attach to
[18:31] AODV-RPL 18:24 - 18:33 (9 min.) Charlie
* changes: some editorial. appendix to describe simulation. Code will be
released. * technical change: instance ID, explanation of S bit. * details
about example topologies added/changed * pick the odd/even bit in the
instance ID to indicate rreq vs. rrep * if RREP turns out to have id of
RREQ + 1, saves putting the TargetAddress in the DIO. If conflict, assign
another id and put the TargetAddress. * Pascal: Seems there should be no
conflict. Name space is from the Requester. * Charlie: need to think about
it. Pascal: if we save this bit and complexity, would be great. * Pascal
crafts a solution: One bit Global/Local. One bit already for the direction.
You don't need the odd/even trick at all.
[18:39] RPL-info 18:33 - 18:38 (5 min.) Michael
* rewrote the security for IP in IP. Need good BCP 38 filtering at the root.
* Maybe problem for with efficient ND in the backbone
* IP in IP causes longer path that can be alleviated by DAO projection
* humm for document approval: humms heard. Against? none heard.
[18:41] Meeting ends