Online decentralized adaptive optimal controller design of CPU utilization for Distributed Real-Time Embedded systems

In large-scale Distributed Real-time Embedded (DRE) systems, the end-to-end tasks contain chains of subtasks distributed on a large number of CPUs. Controlling their CPU utilizations at desired values is one of the most effective ways to ensure system end-to-end deadlines. For these DRE systems, decentralized control is desired to ensure system scalability and global stability. Recently, researchers have proposed solutions based on Model Predictive Control (MPC) for the decentralized utilization control problem. Although these approaches can handle a limited range of execution time estimation errors, the underlying DRE systems may suffer performance deterioration or even become unstable when large estimation errors exist in real systems. In this paper, we propose a new decentralized optimal controller design for CPU utilization to address this problem. The approach leverages Recursive Least Square (RLS) for adaptive model identification and uses Linear Quadratic (LQ) optimal controller for online tasks' execution rates adjustment. Simulation results demonstrate the proposed approach can ensure good system performance even when large constant or varying execution time estimation errors exist.