A study on characterizing energy, latency and security for Intrusion Detection Systems on heterogeneous embedded platforms

Drone swarms are increasingly being used for critical missions and need to be protected against malicious users. Intrusion Detection Systems (IDS) are used to analyze network traffic in order to detect possible threats. Modern IDSs rely on machine learning models for this purpose. Optimizing the execution of resource-hungry IDS algorithms on resource-constrained drone devices, in terms of energy consumption, response time, memory footprint and guaranteed level of security, allows to extend the duration of missions. In addition, the embedded platforms used in drones often incorporate heterogeneous computing platforms on which IDSs could be executed. In this paper, we present a methodology and results about characterizing the execution of different IDS models on various processing elements, namely, Central Processing Units (CPU), Graphical Processing Units (GPU), Deep Learning Accelerators (DLA) and Field-Programmable Gate Array (FPGA). In effect, drones operate in different mission contexts in terms of criticality level, energy and memory budgets, and traffic load, so it is important to identify which IDS model to run on which processing element in a given context. For this sake, we evaluated several metrics on different platforms: energy and resource consumption, accuracy for malicious traffic detection and response time. Different models, namely Random Forests (RF), Convolutional Neural Networks (CNN) and Dense Neural Networks (DNN), have been implemented and characterized on different processing elements/platforms. This study has shown that relating the chosen implementation to the resources available on the drone is a judicious strategy to work on. It highlights the disparity between IDS implementations characteristics. For example, the inference time ranges from $1.27\mu s$ to 30 ms, the energy consumption per inference is between $10.7\mu J$ and 70 mJ, and the accuracy of the IDS models is between 65.73% and 81.59%. In addition, we develop a set of guidelines for choosing the best IDS model given a mission context.