Adapting to the Flow: Reinforcement Learning for Dynamic Priority Assignment in TSN

Real-time systems employ prioritization schemes to
accommodate different traffic classes with specific quality of
service (QoS) requirements. However, in some scenarios where
numerous high-priority packages are transmitted, lower-priority
packages may fail to meet their deadlines, leading to a significant
decline in scheduling performance. Sending high-priority flows
excessively early does not provide any additional benefits beyond
meeting the deadline. Instead, it is more effective to utilize
this buffer time for lower-priority traffic and ensure on-time
transmission of high-priority traffic. We propose an adaptive
dynamic priority assignment scheme that utilizes reinforcement
learning (RL) to address this issue. This enables adaptation
to changing network conditions and continual improvement in
performance over time. Additionally, we present and discuss two
potential configuration candidates that can be utilized within the
proposed scheme.