FIRM REAL-TIME SYSTEMS
Al Mok

We are at a juncture in history when the convergence of computer and telecommunications technologies is transforming human civilization in a fundamental way, and real-time computing technology lies in the heart of this ongoing transformation. Real-time systems are computer-driven systems which must respond to the environment in a timely fashion. It has been observed and vigorously argued by our industrial colleagues that extant real-time systems technology lacks the adaptivity needed by upcoming applications on the internet. While I concur with this observation, I also think that much of the foundation on which to build adaptive real-time systems will be derived from our understanding of *firm* real-time systems technology.

To put this issue in perspective, consider the constituents of traditional hard-real-time systems: (1) the knowledge content which usually consists of a set of state variables of the physical system under control, (2) the real-time performance constraints which is usually expressed in terms of sampling rates and deadlines etc., (3) the resource management technology, e.g., RMA, cyclic executives etc. It is not hard to project the corresponding constituents of the next-generation real-time systems: (1') the knowledge base will not be homogeneous and may consist of facts, pictures, tables etc., (2') the real-time performance constraints which are more forgiving such as quality of service (QOS) requirements, (3') the resource management technology which is very much an open research question.

There are two schools of thought for determining (3'): (A) the best-effort providers and (B) the guarantee providers. Unlike hard-real-time systems, economic considerations will probably lead designers to gravitate toward (A) and away from (B). While there is little arguing that the hard-real-time approach is too conservative in view of (2'), it should be recognized that adaptivity must be designed in, if the system is to be able to cater to different customer needs. Best-effort providers are not effective when there is a price differential between different levels of QOS. Even in military applications, one can envision multiple and possibly dynamically determined priorities for different services and end-users. To take into consideration the price differential necessarily means that providers have to set limits on the efforts in satisfying different customer needs. This is indeed the essence of adaptivity in real-time systems: being able to adjust the level of QOS guarantees in response to price differentials.

A firm real-time system is one which is committed to meet QOS requirements on contractual terms. Relative to a set "price" (which may not be monetary) a set of QOS requirements will have to be met as if they are hard requirements. However, the price can be negotiable, and may in fact be collected subject to the actual QOS provided. The research problem is to establish the quantitative relationship between QOS and price, and this obviously bears on extant theory of hard-real-time systems which can be used, for example, to determine admission control policies. Because of (1') and the evolution of mulimedia technology, theories for designing firm real-time systems must encompass different quality measures of data. Notions such as imprecise computation and similarity will play a central role in this exciting area.