Runtime environments for IoT data processing systems based on the actor model often apply a thread pool to serve data streams. In this paper, we propose an approach based on Reinforcement Learning (RL) to find a trade-off between the resource (thread pool in server machines) usage and the quality of service for data streams. We compare our approach and the Thread Pool Executor of Akka, an open-source software toolkit. Simulation results show that our approach outperforms ThreadPoolExecutor with the timeout rule when the thread start times are not negligible. Furthermore, the tuning of our approach is not tedious as the application of the timeout rule requires.

基于 actor 模型的 IoT 数据处理系统的运行时环境通常应用线程池来为数据流提供服务。在本文中，我们提出了一种基于强化学习 (RL) 的方法，以在资源（服务器机器中的线程池）使用与数据流的服务质量之间进行权衡。我们将我们的方法与开源软件工具包 Akka 的线程池执行器进行了比较。仿真结果表明，当线程启动时间不可忽略时，我们的方法在超时规则方面优于ThreadPoolExecutor。此外，我们的方法的调整并不像超时规则的应用所要求的那样乏味。