Guaranteeing Data Freshness in Distributed Real-Time Systems
Abstract
Control systems for modern large-scale military equipment such as battleships are deployed on widely distributed computing platforms, composed of multiple processing nodes connected by a communication network. In such control systems, readings from multiple sensors are processed by a series of tasks, until ultimately being transformed into actuator commands. To achieve desired quality of control, it is necessary that actuator commands are derived from fresh data; that is, sensor readings and user inputs used in computing the actuator commands are sufficiently recent and adequately represent the state of the system and its environment. Freshness requirements depend on system dynamics are supplied by control engineers. In order to satisfy freshness constraints, computing and communication infrastructure needs to be appropriately designed. In this proposal, we aim to develop foundational techniques to perform freshness-aware design of computation and communication infrastructure used in the implementation of distributed control systems. We concentrate on the following parameters of the infrastructure that have direct impact on propagation of data through the system: (1) periods of real-time tasks that perform data processing steps; (2) allocation of real-time tasks to computational nodes on the platform; (3) scheduling algorithms used to schedule real-time tasks on each of the nodes; and (4) communication schedules for the transmission of time-sensitive data between tasks.We will develop algorithms to adjust these parameters system-wide in order to guarantee that end-to-end freshness constraints are satisfied along all data flow paths from data sources (such as sensors) to data sinks (such as actuators and operator displays). In addition, they will also minimize the overall computation and communication load imposed by control system tasks onthe execution platform, formulating the task of parameter selection as an optimization problem. We will build on the preliminary work we have performed earlier, where we consider a similar problem on a uniprocessor system. Work proposed for this project addresses several challenges that arise from a centralized to a distributed setting, namely allocation of control tasks andscheduling of data transmissions between processing nodes. We will structure the proposed research as a series of work packages, each of which will consider a problem statement of increasing complexity. We will start with a fixed allocation of control tasks to nodes, so that only the communication latency needs to be considered. Subsequent research will also address the problems of deciding optimal task allocation and generating a suitable communicationschedule. These extensions will make the problems much more complex. To counter this complexity and allow our algorithms to scale to practical problems, we will explore a cooptimization approach, where parameters of different subsystems (such as control task periods vs. their allocation to processing nodes) are derived in an iterative manner. Research efforts will be complemented by an extensive evaluation that will run throughout the duration of the project and will guide research by identifying bottlenecks in the algorithms early in the research process. The team consists of three investigators with complementary backgrounds and expertise and a long history and a clear track record of working together on large-scale projects. All investigators are leaders in the real-time systems community for their work related to providingtiming guarantees for complex systems. The team has performed substantial preliminary work, including task allocation and communication schedule generation. Outcomes of the research will have immediate effect on the practice of design and engineering of large-scale control systems. We expect that results obtained in this research will be applicable to systems relevant to the ONR s mission.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 26, 2019
- Source ID
- N000142012744
Entities
People
- Insup Lee
Organizations
- Office of Naval Research
- United States Navy
- University of Pennsylvania