350 likes | 672 Vues
RPL (pronounced ripple) Routing Protocol for Low Power and Lossy Networks Interim Meeting . draft-ietf-roll-rpl-02.txt Tim Winter & Pascal Thubert, editors. RPL Status. WG draft draft - ietf -roll- rpl -02 DT was dissolved. Thanks DT! Added a number of feature Too many? Too complex?
E N D
RPL (pronounced ripple)Routing Protocol for Low Power and Lossy NetworksInterim Meeting draft-ietf-roll-rpl-02.txt Tim Winter & Pascal Thubert, editors IETF 75 – Roll WG – July 2009
RPL Status • WG draft draft-ietf-roll-rpl-02 • DT was dissolved. Thanks DT! • Added a number of feature • Too many? Too complex? • Requires • Re-organisation to avoid dups. • FSM details, description from a node standpoint • Hotest issues • Loop avoidance and detection – tradeoff & mechanisms • P2P routes, DAO fan out, DIO prefixes • MultiDAG, topologies and DV between DAGs IETF 75 – Roll WG – July 2009
Implementer feedback • Dev has started • 4 candidates, • multiple environments • Traces available • Thanks Thomas and Phil • But complain that • Spec too big • Protocol too complex Interim meeting – Roll WG – Sep. 2009
Draft update What did we add recently? IETF 75 – Roll WG – July 2009
OCP0 • OCP 0 is fully spelled out • DAG towards the nearest exit • Uses abstract rank metric • Based on the decimal rank • Add OCP1 for floating? • Other OCPs => in the metrics draft Interim meeting – Roll WG – Sep. 2009
Multicast DAO • NA DOA can now be sent to all nodes • Allows P2P line of sight • Between nodes (not necessarily routers) • MUST NOT be used to redistribute info • Only about stuff owned by this node • Q: redistrib in the unicast NA-DAO? • Not done at the moment • Doable but probable additional complexity Interim meeting – Roll WG – Sep. 2009
Discussions on the list What do we want to achieve IETF 75 – Roll WG – July 2009
Terminology • Need terms for next hops • selectable parents • Non selectable parents • Siblings Interim meeting – Roll WG – Sep. 2009
Floating DAG • What are they for ? • Can be transient for network reconf • Can be permanent as an isolated network • What about OCP? • OCP0 is not ideal for the permanent case • Needs OCP1 that builds a wider DAG for any to any • Can benefit from minimum preference setting to replace the grounded bit. Interim meeting – Roll WG – Sep. 2009
Sibling Loops • Sibling is a necessary evil • Improves FWD diversity • Parents MUST be preferred • No absolute detection • Difficult when multiple sibling hops • But that’s not necessarily so useful • Current preventive rules: • Aggressive TTL decrement, parent preference • No Reverse Interim meeting – Roll WG – Sep. 2009
Missing ~DAO Loops • Comes from DAO states not in sync • A parent has a DAO route but the child has none • Many more or less conplex ways to get there • Detected by simple RPF check. • Proposed preventive rule: A node must drop a packet from a parent if it would route it to a parent (or sibling) Interim meeting – Roll WG – Sep. 2009
Reconfiguration loops • This is when a parent jumps in its subDAG • Risk is reduced by detaching the DAG • But imperfectly, based in unreliable mcast • Detection is required • Can be fully prevented by the seq counter • Additional complexity and delay • Looking for alternate / simplification • DAG merge and split is necessary anyway • A great detection could allow more risky Interim meeting – Roll WG – Sep. 2009
Feature evolution Have we gone too far? A brief history of recent evolution IETF 75 – Roll WG – July 2009
DAG vs. tree • Idea: • Accept multiple parents and siblings • Perceived Benefits • Multiple path inwards • Reality check • More complex implementation (eg sequence) • More states. Metrics aggregation. • Necessary evil for many link types Interim meeting – Roll WG – Sep. 2009
Sequence Number • Brings true loop avoidance • Allows moving down within same DAG • Complexity: • When to trigger a new sequence • What to do of parent with different sequences • How often is the sequence flooded Interim meeting – Roll WG – Sep. 2009
Rank range • Idea: • replace the simple depth incremented by 1 • with a more complex metric with decimal increment • Perceived Benefits • Better mapping with real metrics • Better resulting DAG • Reality check • Broken text and greediness rules • Unpreferable parents. Harder to get siblings • Still discuss on the range. Large enough? Interim meeting – Roll WG – Sep. 2009
Source Route • Idea: • Using source route along the DAG • Perceived Benefits • Save DAO intermediate states • Reality check • No path recovery • Risk of saturation at the root • Need new DIO field / complexity • Another necessary evil for certain node types Interim meeting – Roll WG – Sep. 2009
DAO fan out • Idea: • Send DAO to more than one parent • Perceived Benefits • Multiple path outwards • Reality check • DAG properties are asymmetrical • Fan out not solved yet • Broken for source route Interim meeting – Roll WG – Sep. 2009
MultiDAG • Idea: • Belong to multiple DAGs • Perceived Benefits • More routes, constrained routes • Reality check • More complex DIOs and flows • Eg: when does parent leave a DAG • Distance comparison between DAGS • We need to draw a line somewhere Interim meeting – Roll WG – Sep. 2009
Simplifying RPL Simpler is better When it’s REALLY simpler IETF 75 – Roll WG – July 2009
Sibling loops See Dominique’s presentation Interim meeting – Roll WG – Sep. 2009
Multiple DAGs See Thomas’ Mail Simplifying RPL: (1) multiple DAGs Suggest support to mDAG Echoing Richard Interim meeting – Roll WG – Sep. 2009
Relax loop avoidance • Can we avoid separating the DAG? • Proposal on the list: • allow DAG inconsistency • Use RA-DIO + trickle to fix that quickly • But then, back to RIP and count to infinity • Also, what about a better detection? Interim meeting – Roll WG – Sep. 2009
Lowering transient loss • A node has to detach when no parent • Would allow a child to follow regardless • But Children can’t follow rank increase • So the node could actually move down • To maintain its own connectivity Interim meeting – Roll WG – Sep. 2009
Oustanding issues Things we might need to consider… Or not? IETF 75 – Roll WG – July 2009
Asymmetrical links • Problem: • Best inwards path might be poor outwards • Proposed approach • Diff preferred parents inwards and outwards • Compute Rank as the worst of the 2 • Impact • Probably limited to OCP Interim meeting – Roll WG – Sep. 2009
Security ? Interim meeting – Roll WG – Sep. 2009
More P2P Can forking DAGs suffice? Interim meeting – Roll WG – Sep. 2009
Mobile nodes • Problem: • Find a DAG that reach Home • Default route via the DAG or Home? • Proposed approach • Tie the movement detection to RPL • Hold down routers that do not reach home • Impact to RPL • Probably limited to OCP Interim meeting – Roll WG – Sep. 2009
Compression In particular for source routing Labels? Interim meeting – Roll WG – Sep. 2009