Home > Colloquia > Monday, April 27, 2009

Real-time data from sensors are the first-class citizens in cyber-physical systems (CPS) since they interconnect cyber elements and physical elements of CPS. A real-time data service, which maintains the freshness of data while satisfying other key performance goals such as timeliness of transactions and small resource consumption, is an essential architectural component in many CPS applications. In this dissertation, architecture and algorithms to accomplish such real-time data services are presented in a stepwise fashion.

 

First, a mechanism to make systematic trade-offs between transaction timeliness and data freshness for data-intensive CPS applications is discussed. The proposed solution does not require that the entire data reside in main-memory; hence, it has much broader application for CPS than previous real-time database approaches. To satisfy multi-dimensional performance goals, such as transaction timeliness and energy consumption, a power-aware buffer cache management algorithm, which exploits a Multiple Inputs/Multiple Outputs control structure, is developed and the characteristics of flash storage devices are considered. For efficient fusion of distributed sensor data, we present

PRIDE (Predictive Replication In Distributed Embedded systems) that provides a global view on underlying physical processes to all participating embedded devices. The model-driven approach of PRIDE not only reduces the communication cost of potentially massive amounts of sensor data, but also provides an accurate estimation of the true states of physical processes. Finally, DRACON (Decentralized Replication And CONtrol)’s replica-sharing and decentralized feedback control mechanisms provide a scalable and timely real-time data access mechanism when a large number of small-scale CPS are inter-connected in Internet-like wide area networks.

 

The proposed work is evaluated using several methods: analysis, simulation, and full embedded implementation. Results from experiments show that the proposed adaptive solutions provide robust and controlled behavior in terms of guaranteeing the desired performance under highly unpredictable cyber/physical environments.

Biography: