Towards an Architecture for Distributed Cyber-Physical Systems

Tarek Abdelzaher, Associate Professor, UIUC, Biography link: http://www.cs.uiuc.edu/homes/zaher Introduction An expanding frontier for computer scientists lies at the intersection of the logical and physical realms. This direction is motivated by three fundamental trends that have great implications on the future of computer science as a discipline. These trends are: • Moore’s law. It implies increased miniaturization and reduced cost of hardware, leading to its gradual proliferation. • The widening human/machine bandwidth gap. While computing becomes faster and more pervasive per Moore’s law, the human information processing capability and human population evolve much slower. Over time, this leads to an increasing gap between the ability of computing devices to collect information and the ability of humans to consume it. Therefore, human participation in information collection and processing will become increasingly more marginal and at an increasingly higher (supervisory) levels of abstraction. Computing devices will collectively need to become more autonomous to avoid the human bottleneck. This implies that they will need to have their own means of interaction with their environment with progressively less human mediation. In other words, not only will computing be more autonomous but also increasingly more embedded. • The cost of “lack of communication.” There will always be a cost to the lack of communication or knowledge. For example, the quality of optimization and decision making can always be improved by interconnecting relevant information sources. This leads to a fundamental tendency for (i) global interconnection of the proliferating future autonomous, embedded devices and (ii) reducing their communication cost. The above three fundamental trends herald the emergence of omnipresent, increasingly autonomous, globally interconnected networks of embedded devices with their own means of interaction with the physical world. Applications may include (i) disaster response networks, (ii) complex real-time control networks (such as power-grid control), (iii) environmental sensing networks to explore environmental phenomena at an unprecedented spatial and temporal granularity, and (iv) personal sensor networks. A fundamental question that arises is one of architecture. Namely, what are the reusable middleware components, network protocols, and system services, at different levels of abstraction, that can form a layered, distributed software architecture for cyber-physical computing? Is the existing networking infrastructure sufficient? Are current distributed middleware frameworks (such as web services) a good match to the needs of cyber-physical applications? To answer these question, we explore the deficiencies of current software systems and networks (when applied to cyber physical systems), highlighting new challenges that emerge in the envisioned future.