This paper presents a method intended to estimate the longest and the shortest possible response times in a real-time system. The system is assumed to have a hierarchical level structure where the processes are activated using a preemptive priority schedule.
The method essentially proceeds in a top-down manner through the levels of the system. The processing-time needs at each level are estimated and, using waiting-time estimates, are transformed level by level to the longest (and the shortest) possible real-time needs.
This paper is a shortened version of a longer report, where the results were derived using the methods of temporal logic. It is interesting that temporal logic can be used even in this kind of performance-related work. On the other hand, it is not clear that the paper has benefited from this background. The methods of temporal logic are obviously not necessary to derive the results. On the contrary, it is quite possible that a conceptual model based on queueing theory could have served as a more natural framework for the presentation. After all, in performance evaluation the importance of certain modeling details is not so much a logical as a physical question.
The results are rather straightforward, but the level-oriented analysis might interest a reader unfamiliar with the analysis of queues with priority scheduling. The applicability of the results is not quite clear. First, the shortest and the longest possible “sleeping times” of every process are essential for the analysis, and these values are not always available in a real-time environment. Second, the applied worst-case analysis is far too pessimistic in some environments, for example, in cases with many independent low-frequency events. On the other hand, when worst-case analysis is really needed, the method could prove useful.
The presentation of the paper is adequate, and it is relatively easy to read.