Programming Support for Distributed Optimization and Control in Cyber-Physical Systems

Abstract

Large-scale actuator control problems in Cyber-Physical Systems (CPSs) are often expressed within the networked optimization model. While significant advances have taken place in optimization techniques, their widespread adoption in practical implementations is impeded by the complexity of inter-node coordination and lack of programming support that is necessary for sharing information coherently between distributed and concurrent controller processes. In this paper, we propose a distributed shared memory (DSM) architecture that abstracts away the details of inter-node coordination from the programmer resulting in simplified application design. It maintains data coherency through explicit use of mutual exclusion lock primitives that serialize access to coarse subsets of shared variables using fine-grained read/write permissions. The underlying lock protocol is deadlock-free, fair and safe, and reduces response time and message cost by 81.6% and 72.8% respectively over a conventional DSM implementation with coarse access permissions. Moreover, in a representative application example, the proposed framework reduces application code size by 76% and total latency by 22% over a hand-crafted implementation.