A Map-Reduce Parallel Approach to Automatic Synthesis of Control Software
This work addresses the computational bottleneck of control software synthesis for large-scale systems, offering a scalable parallel solution.
The authors present a Map-Reduce parallel algorithm for automatic synthesis of control software for discrete-time linear hybrid systems, achieving over 65% efficiency and reducing synthesis time from 25 days to 16 hours on a 60-processor cluster for an inverted pendulum problem.
Many Control Systems are indeed Software Based Control Systems, i.e. control systems whose controller consists of control software running on a microcontroller device. This motivates investigation on Formal Model Based Design approaches for automatic synthesis of control software. Available algorithms and tools (e.g., QKS) may require weeks or even months of computation to synthesize control software for large-size systems. This motivates search for parallel algorithms for control software synthesis. In this paper, we present a Map-Reduce style parallel algorithm for control software synthesis when the controlled system (plant) is modeled as discrete time linear hybrid system. Furthermore we present an MPI-based implementation PQKS of our algorithm. To the best of our knowledge, this is the first parallel approach for control software synthesis. We experimentally show effectiveness of PQKS on two classical control synthesis problems: the inverted pendulum and the multi-input buck DC/DC converter. Experiments show that PQKS efficiency is above 65%. As an example, PQKS requires about 16 hours to complete the synthesis of control software for the pendulum on a cluster with 60 processors, instead of the 25 days needed by the sequential algorithm in QKS.