Performance Impact of Failure and Recovery Mechanisms in LoRa Sensors

Abstract

LoRa sensor deployments are widely used in monitoring applications, where multiple devices generate packets continuously or periodically. In these scenarios, system performance is affected not only by the number of sensors and the packet generation rate, but also by intermittent failures. When a sensor becomes unavailable, packets may be lost and throughput decreases. The recovery time then defines how long the system remains in this degraded condition. Traditional long-term analyses may hide these short-term effects, especially the sudden increase in packet losses during and shortly after failures. This work proposes an analytical model based on Stochastic Petri Nets to represent the behavior of LoRa sensors under failure and recovery conditions. The model combines the sensor availability cycle, alternating between operational and failed states, with the packet flow, including packet generation, waiting, and transmission. From this model, it is possible to estimate throughput, the number of delivered packets, and packet losses over time. The model was validated against the IoT-FailTolerantSim simulator using 30 replications and statistical tests. The results showed good agreement for packet arrival rates from 1 to 7 packets per hour. For 8 packets per hour, the model still presented a limited deviation of approximately 4.3%. Four scenarios were evaluated to understand the individual effects of packet arrival rate, repair time, failure interval, and transmission time on delivery performance. The packet arrival rate determines how quickly packets accumulate and are lost during a failure. The repair time defines how long the degraded period lasts. The failure interval affects how often these degraded periods occur. Finally, the transmission time directly reduces throughput, although it does not change the number of packets lost during failures. The results show that tripling the load increases losses by 2.7 times within the same failure window. Increasing the repair time from 100 h to 300 h reduces the total number of delivered packets by approximately 22% over 1000 h of operation. In addition, quadrupling the per-packet Time-on-Air (ToA) from 0.5 h to 2 h reduces effective throughput by approximately 64%. These findings provide useful guidance for planning maintenance strategies and defining service requirements in critical LoRa deployments.