Date of this Version
To date, research in reasoning about timing properties of real-time programs has considered specification and implementation as separate issues. Specification uses formal methods; it abstracts out program execution, defining a specification that is independent of any machine-specific details (see [I, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] for examples). In this manner, it describes only the high-level timing requirements of processes in the system, and dependencies between them. One then typically attempts to prove the mutual consistency of these timing constraints, or to determine whether the constraints maintain a safety property critical to system correctness. However, since the model has abstracted out machine-specific details, these correctness proofs either assume very optimistic operating environment (such as a one to one assignment of processes to processors), or make very pessimistic assumptions (such as that all interleavings of process executions are possible). Since neither of these assumptions will hold in practice, these "predictions" about the behavior of the system may not be accurate.
The implementation level captures this operating environment: a real- time system is characterized by such things as process schedulers, devices and local clocks. However, advances here have been primarily in scheduling theory (examples of which are [15, 16]) and language design (examples of which are [15, 16, 17, 18,19,20]). Unfortunately, since formal models have not been used at this level, proofs of time-related properties cannot be made. To construct these proofs, we must show that an implementation is correct with respect to a specification; timing properties that can be shown to hold about the specification will therefore be known to hold for the implementation. We therefore need to represent the implementation formally so as to prove that the implementation satisfies the specification. The proof of satisfaction requires a well-defined formal mapping between the implementation and specification models.
We therefore propose to develop an integrated bi-level approach to the problem of reasoning about timing properties of real-time programs. At the specification level, we will use the Timed Acceptances model, a logically sound and complete axiom system which we have recently developed . Using this model, the effect of interaction among time dependent processes can be precisely specified and then analyzed. We will then develop a formal implementation model (similar to the specification model) which captures operational behaviors: for example, the assignment of processes to processors, assumptions about scheduling and clock synchronization, and the different treatment of execution and wait times. A mapping will then be formulated between these two layers.
The bulk of our proposed work will be to formulate the implementation layer and define a mapping between it and the specification layer. We also need to continue work on the Timed Acceptances model to facilitate its use as a specification model, and to provide "hooks" for mappings between the two layers.
The rest of this proposal is organized as follows. The next section overviews related work in formal specification models. Section 3 describes our current specification model and proposed enhancements. We also detail the proposed implementation model, and required properties of the mappings between the two models. Section 4 provides a summary of the proposed research, and a yearly plan.
Insup Lee, Susan B. Davidson, and Richard Gerber, "Formally Integrating Real-Time Specification: A Research Proposal", . September 1988.
Date Posted: 05 December 2007