Back to Advance Program|
Living with uncertainty: A systems architect's view on real-time
Sjir van Loo, Principal systems architect, responsible for the Philips Research long-term objectives for the Embedded Software TechnologyTo appear at Euromicro Conference on Real-Time Systems (ECRTS03), Porto, Portugal, 2-4 July 2003
BioSjir van Loo (1949) graduated in physics at Eindhoven University of Technology in 1974. He joined Philips in 1985, and Philips Research Laboratories in 1992.
He has been working as a software and systems architect for over 25 years, and was involved in the design and realization of many industrial and research systems, in the professional as well as in the consumer domain. These systems include Remote Surveillance systems, Distributed Real-time Operating systems, MRI scanners, Video-on-Demand servers, mobile phones, digital set-top boxes and TV-sets. His current research interests are systems architectures for Ambient Intelligence, resource management for low-cost high-volume electronics and programming paradigms for newly emerging programmable silicon architectures. In his current position he is also responsible for the Philips Research long-term objectives for the Embedded Software Technology programme. Furthermore, he is the manager and a co-founder of the Philips Research Systems Architecting training programme, established in 1997.
He gives regular lectures on systems architecting and design, among others in post-masters courses, both inside and outside of Philips. He is a member of the advisory board of the Information Technology Institute at the Fontys University of Professional Education in Eindhoven.
AbstractOver the last 10 years the consumer electronics industry is driven by the digitisation and compression of audio, e.g. CD and MP3, speech, e.g. GSM phone, and more recently video, e.g. DVD and Digital Video Broadcast. Increasingly, media processing in these devices will become programmable. In fact, the whole spectrum from large dedicated function blocks, like an MPEG2 decoder, up to general-purpose processors is used and will be used in media processing. Which alternative to choose from this spectrum, depends on architectural trade-offs, considering aspects such as functionality, flexibility, cost, design complexity and power consumption, to name a few.
Consumers expect high quality audio and video from their devices, without any hick-ups or artefacts, not to mention image stalls or even system crashes. In terms of realization, this implies that a consumer device like a digital TV set is a real-time system with hard deadlines. Traditionally, the real-time community deals with such deadlines by analysing the system from a worst-case perspective, and implementing the system based on a model of underlying hardware that exhibits predictable (worst-case) behaviour.
In future consumer systems the systems architect, responsible for the design of the system as a whole, will be confronted with three sources of uncertainty, which, if not dealt with properly, will result in highly unpredictable system behaviour. The first source of uncertainty is induced by the application, caused for instance by large variations in the complexity of the incoming or outgoing data-streams. This variation makes it impractical or even impossible to define useful worst-case execution times for processing the stream. A second source of uncertainty is induced by technology. External causes, like heat dissipation, power management or radiation, might cause an unpredictable change in the availability or performance of resources, such as the amount of cpu-cycles or bus bandwidth. The last source of uncertainty, architectural uncertainty, we introduce ourselves when designing systems in which we do not control interference of activities in shared resources such as memory, caches and busses. This kind of uncertainty exhibits in general very non-linear behaviour, and prevents simple composition of otherwise correct functionality.
It is obvious that in consumer devices, where cost is a major issue, and where we want the available silicon to be used as effectively and efficiently as possible, a worst-case approach is not realistic. We will have to live with unavoidable uncertainty, and deal with it in such a way that we maintain the high quality and robustness consumers are used to.
Achieving this is a major architectural effort that involves the application research and design community, the silicon research and design community, as well as the real-time research and design community. A concerted, systems architectural approach is required, to assure a coherent system that in the end exhibits the desired behaviour and fulfils the purpose of the system in all functional as well as non-functional aspects.
In my talk I will elaborate on the uncertainty issues mentioned above, address the research and engineering challenges posed onto the real-time community, and present my view, a systems architect's view, on the way to move forward into a world in which we deal with uncertainty rather than ignore it and hope for the best.
14 Oct 2003 at 17:14:26