PRIMO: Propagation-Indexed RPL Mobility Observations

This version is a supplement. It holds one file. The complete dataset is version 1 plus version 2. PRIMO comprises twenty archives totalling 47 GB. A Zenodo record is limited to 50 GB, and the twenty archives could not be fitted into a single record together with the headroom that a deposit needs while it is being assembled. The dataset was therefore split across two versions of the same concept record: Version 1 (10.5281/zenodo.21717866) — nineteen archives, 45.3 GB: all six transmit powers of armor_paper and of rl, five of the six of rpl, plus primo-inputs.tar.gz and primo-toolchain.tar.gz. Version 2 (this record) — one archive, 1.6 GB: primo-rpl-p8.tar.gz, the remaining cell of the grid, standard RPL at −8 dBm. Both versions are required. Resolving the concept DOI 10.5281/zenodo.21717865 leads to the most recent version, which is this one; version 1 must be downloaded separately for the other nineteen archives. Neither version is a superset of the other, and version 2 does not supersede version 1. The split is an artefact of the storage limit, not of the experimental design: the grid is factorial and complete, and primo-rpl-p8.tar.gz is one cell of it, in the same format and produced by the same run of the same campaign as the other nineteen. Should a quota increase be granted, a version 3 carrying all twenty archives in one record will be deposited, and this note will say so. A factorial simulation dataset of RPL routing under node mobility, designed so that the radio propagation environment is the only thing that changes between comparable cells. What the dataset answers Mobility-aware objective functions extend RPL with constants that bound how long a parent remains usable. Those constants are normally fixed by trial and error on a single radio configuration, which leaves open how far a value calibrated in one environment travels to another. Separating that dependence from other sources of variation requires a campaign in which the radio changes and nothing else does. PRIMO is that campaign. Design The grid crosses 13 propagation environments with 6 transmit powers, 8 mobility scenarios, 10 random seeds and 3 routing models: 234 cells and 18,720 simulation runs in total. Propagation. Nine log-distance configurations vary the RSSI inflection Transmit power. 0, −4, −8, −12, −16 and −20 dBm, inside the programmable Scenarios. Four network sizes (21, 33, 41 and 53 nodes: one sink, static Routing models. Standard RPL with MRHOF as the reference floor; a Each run covers 3,600 s of simulated time; the first 900 s are marked as Why the mobility is invariant The 13 environments are clones of a single mobility base. The trajectory files are hard linked, not copied, so every environment references the same inode and the geometry is provably identical across the propagation sweep. Any difference measured between two environments is propagation, never movement. This is verifiable in the release with ls -i. What is released Aggregated metrics (armor_metrics.csv), one file per cell: delivery Raw per-node counters (per_node.csv): role, packets generated and Compressed simulator logs (.log.gz), the complete raw output of every Per-run provenance (.meta.json): the patched simulation file, the Toolchain: the firmware sources, the campaign scripts and the analysis Every node emits a boot-time configuration stamp carrying the routing mode, transmit power, constant values, path-loss exponent and learner state; the extraction step propagates it as columns of every CSV row, so any released number traces back to the binary that produced it without holding the log. Size 49 GB with the compressed logs; 633 MB for the CSV and metadata alone. Reproducibility The firmware builds against an unmodified Contiki-NG tree at commit f86bff9bb, verified by reverting every local modification and rebuilding the three mote binaries. The release does not depend on a private fork of the operating system. Simulation is performed in Cooja with a log-distance radio medium; the mobility base can be regenerated from the BonnMotion command lines included with each scenario. A first reading Averaged over the grid, ARMOR raises the delivery ratio of the mobile nodes from 27.8% to 33.7% and reachability from 69.2% to 94.1% against standard RPL, at the cost of 2.2 times more parent switches. The bandit variant does not improve on ARMOR with fixed constants: the per-window reward does not separate the candidate values, so the adaptation adds variance rather than accuracy. The dataset is released to let that question, and others, be re-examined on the raw counters rather than on published summaries. Citation If you use PRIMO, please cite the accompanying data descriptor.