一种基于强化学习的嵌入式系统抗拒绝服务攻击的缓存调度方案

在多核嵌入式操作系统中,中央处理器对共享最后一级缓存(Last Level Cache,LLC)的资源调度决定了各用户进程的指令周期数(Instructions Per Cycle,IPC),以及对拒绝服务(Denial-of-Service,DoS)攻击的鲁棒性。但是,现有缓存调度方案依赖于具体的LLC调度模型和DoS攻击模型,使中央处理器难以在不同调度环境中的每个调度周期及时获得用户进程的运行信息。因此,文中提出一种基于强化学习的嵌入式系统LLC调度技术,以抵御拒绝服务攻击。该技术根据用户进程的LLC占用起始位置和终止位置,结合反馈的指令周期数、载入未命中率和存储未命中率等信息,优化LLC的占用位置和占用空间。在动态LLC调度环境下,中央处理器不需要预知DoS攻击模型,即可提高指令周期数并同时降低恶意进程的DoS攻击成功率。在多租户虚拟机共同参与的多核嵌入式操作系统中的仿真结果表明,所提技术可以显著提高指令周期数并降低DoS攻击的成功率。

Reinforcement Learning Based Cache Scheduling Against Denial-of-Service Attacks in Embedded Systems

The sharing last level cache(LLC)scheduling of the central processor determines the instructions per cycle(IPC)of the user processes and the robustness of denial-of-service(DoS)attacks in the multicore embedded operating systems.However,existing scheduling schemes rely on the specific LLC scheduling model and DoS attack model,which makes it difficult for the processor to obtain the running information of the user processes in each scheduling cycle under different scheduling environments.Therefore,this paper proposes a reinforcement learning(RL)based LLC scheduling scheme to against DoS attacks in embedded systems,which optimizes the occupied position and the occupied space based on the measured occupied start and end positions,the previous IPC,load miss rate and store miss rate.The processor can jointly increase the IPC and reduce the success rate of the DoS attack from the malicious process without knowing the DoS attack model in the dynamic LLC scheduling environment.Simulations are implemented on the multicore embedded operating systems where multitenant virtual machines participate together,which show that the proposed scheme can significantly increase the IPC and reduce the success rate of the DoS attack.