1. Slide #1 Performance Evaluation of Routing Protocol for Low Power and Lossy Networks (RPL) draft-tripathi-roll-rpl-simulation-04 IETF Virtual Interim WG Meeting – June 2010 Joydeep Tripathi Jaudelice De Oliveira JP Vasseur

2. Draft Update Version 00 - CDF of hop distance and ETX, routing table size, data and control overhead, for each node, over time for sample topology from RPL draft. Version 01 - December 2009 - CDF of path quality, routing table, data & control overhead, for each node, over time and CDF of time to recover from loss of connectivity, with global repair only, for a gathered topology trace. Version 02 - Local repair and its effect on time to recover from loss of connectivity was added. Version 03 – March 2010 - CDF of path stretch factor and delay added. Building routing implemented. Version 04 – June 2010 - Results on scalability on a very large smart metering network added.

3. Simulation Setup Simulated in Castalia WSN Simulator in Omnet++ framework. Real topology with 45 (home & building routing) / 2442 (smart meter network) nodes. Gathered link traces from deployment used to simulate packet drop probability at each transmission, which varies with time. Telos CC2420 Radio with 240 kbps data rate and 802.15.4 MAC/PHY specification. New DAG sequence number each 10/30/60 minutes. ETX metric used as metric to build the DAG. Data Packet size 127 bytes at physical layer.

4. Assumptions No constraint on the size of routing table, and how much information the node can store. Link quality (PDR) remains constant for 10 minutes, and then it is updated. Random links are brought down during the simulation to reflect realistic network failure condition, modeled as a Poison process. If new path metric is better by less than 20% than old one, new parent selection is avoided to prevent frequent flipping of parent – child structure. Nodes multicast prefixes learned from its sub-DAG only, not from others (siblings, etc.).

5. Simulation Results CDF in %age  Slightly worse than ideal shortest path metric.  Follows almost the ideal path. Comparison of ETX Path cost for shortest path and RPL Comparison of Hop Distance for shortest path and RPL (ETX used as metric)  Sometimes better than shortest path.  Minimum in ETX metric may have a longer path in terms of hop distance.

6. Simulation Results (Contd.)‏ CDF in %age CDF of Delay in %age  Fractional stretch factor is larger for shorter path.  Delay is mostly in order of milliseconds. Fractional Metric Stretch with Ideal shortest Path End to End Latency for different hop count range

7. Simulation Results (Contd.)‏ Number of Packets Number of Control Packets Data and Control Overhead Comparison of Control Overhead  Control overhead nearly 1% for nodes close to sink.  For leaf nodes, both control & data packet amount is small.  Decreases with increase in Global Repair timer period.

8. Simulation Results (Contd.)‏ Comparison of Time without connectivity to a parent, for small Indoor Network

9. Future Direction Goal: Informational RFC Poll WG on making it a Working Group Document Next Steps: Incorporate WG comments/feedback. Are there other scenarios/metrics of interest?