1. Wireless CPS in Process Control:

Source of support: NSF (2015--2016) 

A wireless control system involves feedback control loops between sensors and actuators  through a wireless mesh network.  Sensors  measure process variables, and deliver to a controller through the network. The controller sends control commands to the actuators, which then operate the control and safety components to adjust physical processes so the system's performance is optimized for efficiency and safety.  Industrial CPS make critical demands for  reliable  and  real-time communication between sensors and actuators in order to avoid plant shutdowns or accidents. We are working towards theory and system development for wireless CPS. Our remarkable contributions in this area include the establishment of a novel real-time wireless scheduling theory and a scheduling-control co-design framework for holistic optimization of control performance in wireless control systems. One related paper was Best Paper Award Nominee at RTAS 2012 (a flagship conference on embedded and real-time systems.)

For questions, please contact Prof. Abusayeed Saifullah: saifullah@wayne.edu

2. CapNet (Wireless Networked System for Power Capping):

Source of support: Microsoft   

Modern Internet and enterprise data centers continue to scale to meet the growing cloud computing and storage demands by increasing the number of servers. For this large number of servers, cost-effective management (e.g., power management, deployment checks, health diagnostics, asset management) becomes increasingly important. Today, data center management network is built around a wired communication mechanism that is independent from the primary network infrastructure for redundancy and fault isolation reasons.  Hence, the scalability of data center management network faces great challenge when the server rack configurations become increasingly dense. We are working towards developing wireless management network for large-scale data centers. In particular, we want to exploit low-bandwidth wireless technologies such as those based on IEEE 802.15.4 at the leaf layer of the management network. These networks can be built at very low cost, can self-configure, and are resilient to failures.  Our remarkable contribution towards this goal is the development of CapNet, a real-time wireless sensor network system for data center power capping. CapNet was developed based on IEEE 802.15.4 for TinyOS platform. The related paper received the Best Paper Award at RTSS 2014. RTSS is regarded as the highest venue for real-time systems research. 

For questions, please contact Prof. Abusayeed Saifullah: saifullah@wayne.edu

3. Parallel and Multi-Core Computing:

Source of support: NSF (2023--2027) 

Multi-core processors are becoming mainstream in processor design technology. They can enable  the emerging computation-intensive  real-time CPS (e.g., autonomous vehicles)   with stringent timing constraints that  cannot be met on traditional single-core machines.  To take full advantage of multi-core processing, these systems need to  exploit intra-task parallelism, meaning that a  task which itself is parallelizable should  execute on  multiple cores at the same time. However, most classic results in real-time scheduling concentrate on sequential tasks running on multiple  cores.  While they allow many tasks to  execute simultaneously on a multi-core, they do not allow an individual  task to run any faster on it than on a single-core machine. We are working towards developing real-time theory and systems for exploiting intra-task parallelism  on multi-core. Our remarkable contributions towards this goal include multiple real-time parallel scheduling theories and capacity augmentation analyses for parallel task scheduling on multi-core platform. 

Incorporating Cache Benefit. We have also addressed inter-thread cache benefits on multicore platform by incorporating cache sharing into parallel scheduling. For hard-real time systems, cache memory increases execution time variability, increasing the complexity of timing analysis. Cache-aware co-located scheduling aims to improve schedulability by carefully scheduling threads to share cached values. Cache sharing between threads potentially reduces task execution times and increases schedulability with fewer resources. Antithetically, co-located scheduling may reduce parallelism, decreasing efficiency. Thus, identifying the optimal set of threads to co-locate that minimizes the resources required while ensuring timing constraints is a complex challenge.

4. Energy-Efficiency in Real-Time CPS: 

Source of support: WSU Competitive and Graduate School Research Grant (2018--2020) 

Both energy-efficiency and real-time performance are critical requirements in many embedded systems such as self-driving car, advanced robotic system, and disaster response. A recent study using the Ford Fusion autonomy system has revealed that 41% energy is consumed by the computing platform of a self-driving car. We have addressed real-time scheduling of parallel tasks, each represented by a DAG, while minimizing their CPU energy consumption on various multicore embedded systems. Energy-aware real-time scheduling is challenging due to complicated relation between frequency, energy consumption, and execution. Existing study considered only sequential tasks. Energy-aware real-time parallel scheduling of DAGs is highly challenging due to the dependencies among the vertices as well as among their execution lengths. Our contributions include the first energy-aware parallel scheduling of DAGs. It incorporates an adaptive frequency optimization engine through DVFS (dynamic voltage and frequency scaling) for minimizing CPU energy consumption into the classical real-time scheduling policies and makes them energy-aware. We have implemented the results on ODROID multicore boards and on the Intel Xeon multiprocessor platform.