TOMO: TSCH Objective-Function Mobility Observations

Mobile Internet of Things deployments route their traffic with RPL, in which every node selects a preferred parent and an objective function ranks the candidates. When nodes move, links break before the standard metric reacts, and a family of mobility-aware objective functions answers by predicting the rupture from position or from received signal strength. That family was evaluated under asynchronous medium access, where changing parent costs one control message. Time-Slotted Channel Hopping (TSCH) inverts the premise: a parent change costs a 6top transaction that the scheduling function must complete before the new parent can carry traffic, and the standard profile already ranks candidates by transmission feedback. TOMO is the campaign needed to measure whether the predicted gains survive that inversion: four objective functions compared on a scheduled medium, with every function reading the same radio observation. Design The factorial crosses four objective functions with six node speeds and three transmit powers, over twenty paired trajectories per cell drawn from two mobility models. The full grid holds 1,440 runs across 72 cells, with no failed run. Objective functions (4). cego: expected transmission count, with the standard default step of rank for a candidate never transmitted to. mrhof: the same, with the estimator initialized from the received signal strength through the radio reception curve. ltamv: rank from a predictor of the received signal strength, two exponential filters at 8 and 30 s over 36 symbols and six states. armor: rank from a time to rupture computed from position and velocity (Mohammadsalehi et al., IEEE IoT-J 2021), ported to the scheduled medium. Node speeds (6). 1.00, 5.07, 10.00, 15.25, 20.73 and 30.11 m/s, measured over the generated waypoints. Transmit powers (3). 0, −7.5 and −13.6 dBm, giving effective ranges of 47.4, 19.9 and 9.9 m under the reception model R(P) = 14.0 m × 10(P+10.6)/20. Mobility models (2). Manhattan grid and random walk, ten trajectories each, generated with BonnMotion 3.0.1. Every cell holds 41 nodes in a 100 m square: one sink at the centre, twelve static anchors and twenty-eight mobile nodes. Each run covers 2,500 s of simulated time, 1,000 s of static bootstrap followed by 1,500 s of motion over which every released counter is accumulated. What makes the comparison valid Topology held fixed. Node count, anchor placement, sink position and deployment area are identical in all 72 cells, so the number of reachable candidate parents varies with transmit power while the topology does not. Symmetric reading of the radio. The three signal-strength functions obtain the measurement from a single per-neighbour cache, written on reception of any frame and read through one accessor. Measurement noise of 3 dB standard deviation is applied when the sample is written, so two functions evaluating the same neighbour at the same instant read the identical perturbed value. The propagation model is untouched: delivery is still decided by the true received power, and only the value the protocol stack records is perturbed. The one asymmetry favours the competitor. The position-based function receives exact coordinates and velocities, an idealization no deployed receiver would have. Paired trajectories. The trajectory index is paired across cells, so a contrast between two objective functions is computed trajectory by trajectory rather than between independent samples. Contents Complete simulator output — 72 compressed archives, one per cell, 7.7 GB compressed and 74 GB expanded. Each archive holds the twenty run directories of its cell, with the configuration the simulator consumed and its full output. Distilled counters — 144 CSV files, two per cell, twenty rows each. The delivery view carries application packets generated and delivered, parent changes, synchronization and desynchronization events. The cost view carries 6top frames and transactions, cell installations and removals, control frames by type, receptions lost to interference, drops by cause, and accumulated charge, plus four derived ratios whose terms are both present. Mobility traces — 120 scenario files, twenty per speed level, each carrying node roles, the coordinates of sink and anchors, and the trajectory of every mobile node. Simulator modifications — 21 idempotent patch scripts, each guarded by a configuration switch that defaults to the published behaviour, plus an untouched copy of the published simulator from which the modified tree is rebuilt. Scripts — scenario generation, campaign execution, extraction and paired comparison. Reproducibility The modified simulator is rebuilt from the published source by a single script that applies the patches in a fixed order and fails with an error, rather than a warning, if any anchor does not match or if any patch block appears twice. With every configuration switch at zero the rebuilt tree reproduces the published behaviour. The rebuilt tree was verified against a previously collected baseline and reproduced it digit for digit. Configuration switches are checked in both directions. A switch that should be neutral when disabled must reproduce the reference cell in every metric; a switch that should change behaviour must differ from it. The second check exists because a switch that never fires produces a plausible null result and no error. Archives are verified by a full read before the raw directory is removed. If the archive is missing, empty or unreadable, the raw directory is preserved and the execution log records it, so no cell is discarded without a verified copy. Intended reuse The paired structure supports any contrast between the four functions at fixed speed and power. The cost view supports energy accounting under routing instability, since parent changes, 6top transactions, cell operations and accumulated charge are recorded per run. The archives support quantities the extractors did not distil, among them latency distributions, per-node rather than per-network aggregation, and the temporal placement of parent changes within a run. Two limits bound what the release supports. The power axis varies the number of reachable candidates at a fixed node count, so range and node density are confounded along it. Latency, which a scheduled medium determines through cell placement rather than through parent choice, is present in the archives but not in the distilled files. Software Simulation uses the 6TiSCH simulator of Municio et al. at the published revision. Trajectories come from BonnMotion 3.0.1 under OpenJDK 17. Analysis runs on Python 3.10. Archives are produced with GNU tar and gzip. The data are released under CC BY 4.0; the scripts and simulator patches follow the license of the 6TiSCH simulator they modify.