Optimizing Real-Time Intelligent Traffic Systems with LSTM Forecasting and A* Search: An Evaluation of Hypervisor Schedulers
This research explores an Intelligent Traffic System (ITS) designed for real-time optimal routing using traffic forecasting and an A* search algorithm. Leveraging a pre-trained Long Short-Term Memory (LSTM) neural network, I predict traffic flow based on historical data to inform heuristic functions, ensuring optimal route calculations. The heuristic is constructed to be permissible and consistent by incorporating predicted traffic flow and average speed measurements. The experimental setup involves a messaging virtual machine (VM) and a real-time VM within a Xen hypervisor environment, utilizing Apache Kafka and Apache Flink for data flow and processing. I empirically evaluate the latency performance of the ITS under three different Xen schedulers: RTDS, Credit, and Credit2. My findings indicate that the RTDS scheduler provides superior latency guarantees, making it suitable for applications requiring ultra-low latency, whereas the Credit and Credit2 schedulers offer better median performance. These insights highlight the impact of hypervisor scheduler choice on the efficiency and responsiveness of real-time ITS applications.
