Análise do desempenho de sistemas operacionais de tempo real para dispositivos IoT no contexto Multi-core

Abstract

This work evaluates the performance of two open-source RTOSs, FreeRTOS and NuttX, on the Raspberry Pi Pico (RP2040), comparing single-core and SMP execution. The goal is to assess how internal kernel synchronization decisions impact scalability on low-end multicore microcontrollers. Three complementary scenarios are implemented: isolated kernel primitives (to characterize structural overhead under multicore), integrated workloads exercising typical system services, and a controlled contention scenario in which tasks compete for the same shared resource. Each routine runs in fixed time windows, computing throughput by iteration counting, while peak dynamic memory usage (heap) is recorded as a complementary metric. Results show similar behavior for compute-bound workloads, but clear differences when scheduling and synchronization dominate the cost. Under SMP, FreeRTOS tends to degrade more sharply with frequent critical-section entries, since multicore protection may serialize execution even without explicit contention on the same kernel object. In contrast, NuttX preserves scalability under low contention by benefiting from fast paths in some operations, but gradually loses this advantage as contention increases and full synchronization paths are triggered more often. NuttX also reaches a higher peak heap usage, consistent with running with more system services enabled during the experiments.