Leslie C. Lander

Classification of possibilistic uncertainty and information functions

Abstract A possibility distribution is defined as a sequence (pi), 0 ⩽ pi ⩽ 1, such that maxipi = 1. Joint and marginal distributions are defined in analogy with probability theory; however the operations of minimum and maximum replace multiplication and addition. The classical notion of entropy, as a measure of information associated with a probability distribution has as its counterpart an uncertainty measure associated with a possibility distribution. Higashi and Klir proposed several axioms that an uncertainty function should satisfy. They also proposed one such function — given a non-decreasing sequence (pi), its value is p 1 log n + (p 2 −p 1 ) log (n − 1) + (p 3 −p 2 ) log (n − 2) + … In this paper we give a complete description of uncertainty functions, satisfying those axioms. They are classified by montonically increasing functions from the unit interval into itself which fix end-points. To any such function z(x) there corresponds an uncertainty function z(p 1 ) log n + [,z(p 2 ) − z(p 1 )] log (n − 1) + … We also discuss the role of continuity and deduce uniqueness of the measure under and additional linearity property. Because of the interdependency of the axioms, our results assume only the properties of symmetry, expansibility, monotonicity and additivity.

Bayesian analysis for fault location in homogeneous distributed systems

A simple and practical probabilistic comparison-based model, employing multiple incomplete test concepts, for handling fault location in distributed systems using a Bayesian analysis procedure is proposed. This approach is more practical and complete than previous ones since it does not assume any conditions such as permanently faulty units, complete tests, perfect environments, or non-malicious environments. Fault-free systems are handled without overhead; hence, the test procedure may be used to monitor a functioning system. Given a system S with a specific test graph, the corresponding conditional distribution between the comparison test results (syndrome) and the fault patterns of S can be generated. To avoid the complex global Bayesian estimation process, a simple bitwise Bayesian algorithm is developed for fault location in S, which locates system failures with linear complexity, suitable for hard real-time systems.<<ETX>>

A module‐based conceptual framework for largescale expert systems

Increasing utilization of expert systems in organizations has necessitated development of large‐scale expert systems that deal with complex problem domains requiring handling of large knowledge bases. Adapts the concepts and structures of open information systems to outline a conceptual framework for large‐scale expert system development called an open‐based expert system (OES) architecture. The framework proposes partitioning large‐scale expert systems development into smaller and autonomous expert system modules which communicate through a flexible interface. The expert system modules may be built small enough to achieve reliability, and the open structure of the architecture allows the modules to be incorporated into a larger information system. The OES architecture is implemented on vp‐expert shell, and a prototype is developed to demonstrate its working.

An inference design for fault location in real-time control systems

Abstract We develop a probabilistic comparison-based model and fault-location strategies using statistical inference, namely point and set estimation. Our heuristic distance-based algorithm can efficiently locate the faults with a 1 — α level of confidence. We show how the exponential complexity of the global estimation process for fault location in a real-time system S is simplified using this approach. The complexity of the run-time fault location is O(n) for systems with n units. Fault-free systems can be handled, and there is a high likelihood of distinguishing truly faulty units from those that appear faulty because of the imperfect environment, thus eliminating unnecessary replacements during reconfiguration. Our work is intended to be a practical approach to fault location; it accommodates all the random effects of testing real-time control systems in a general way.

Priority inversion in Ada programs during elaboration

Priority inversion is any situation where low priority tasks are served before higher priority tasks. It is recognized as a serious problem for real-time systems. We have observed priority inversion due to the elaboration order in versions of AIsys I Ada and VAX ~ Ada. This paper presents a rigorous description of Alsys Adas priority inversion during elaboration of library packages; this behavior is specified in a semi-formal partial model of the elaboration algorithm of the Alsys 286 DOS Ada Compiler V4.2, running under PC-DOS Version 3.30 on an IBM PC AT. s The model was deduced from a series of experiments designed to illuminate the full scheduling features of the system. These results may provide useful insight for users of this and other Ada compilers. Similar priority inversion was also observed for VAX Ada running under VMS. K e y w o r d s : Ada, Alsys, context clauses, elaboration, performance model, priority, priority inversion, reverse engineering, simulation, task scheduling I Alsys is a registered trademark of Alsys z VAX and VMS are trademarks of Digital Equipment Corporation IBM, PC AT and PC-DOS are registered trademarks of International Business Machines Corporation COPYRIGHT 1990 BY THE ASSOCIATION FOR COMPUTING MACHINERY, INC. Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and or specific permission. In [9] priority inversion is defined to be any situation where a low priority task holds a resource while a higher priority task is ready to use it. In [4] and [5], the definition of priority inversion, attributed to Lui Sha, is given as any situation where low priority tasks are served before higher priority tasks. Priority inversion is recognized as a serious problem for real-time systems [4]. Priority inversion during the execution of an Ada program may occur in a number of situations, including: In a rendezvous when the accept code is being executed by or on behalf of a low priority task while, after rescheduling, it might be executed by or on behalf of a higher priority task, which is blocked, see, for example, [4]. This inversion is permitted and in many cases forced by the Ads semantics [1] and occurs because of the combination of the sequence of task execution and the structure of the program. Priority inversion may occur in selective wait statements where a high priority task at the head of one entry queue is ignored while a low priority task at the head of another entry queue is accepted. Such inversion is permitted but not forced by the Ada semantics [2]. We report elsewhere [8] that this form of inversion may result from applying a deterministic rather than a random selection algorithm. We have recognized a form of priority inversion during the elaboration of Ada programs with the two Ada compilers we have used. Tasks of low priority are served first by the CPU because high priority tasks have not yet been elaborated. Ada semantics would appear to state that, at the beginning of program execution, the highcst priority tasks should run first. However the tasks in a package may begin to run as soon as that package is elaborated. If the lowest priority tasks are in the first package to be elaborated, they will run first. Our observation occurred during experiments of wider scope described below. The work presented in this paper constitutes a portion of a Wash ing ton Ada S y m p o s i u m P r o c e e d i n g s . June 1990 133 larger project in formal system specification methodologies and performance prediction being carried out in the Department of Computer Science at SUNY Binghamton. As a performance measurement and validation vehicle we are building a Communications Gateway [11, 12] which will multiplex several IBM Token Ring LAN users to the DataSwitch of the SUNY-Binghamton Computer Center. The primary purpose ofthc Gateway is to provide an instrumented multi-tasked Ada system whose actual performance will be compared to that predicted via analysis of specification models of the system (see [3]). The Gateway code will be compiled in Alsys Ada and will run in the Alsys run-time environment. In order to develop a detailed performance model [13] (in our case to be used for simulation analysis of the Gateway behavior), we shall incorporate into the model the activity scheduling algorithm associated with the Alsys compilation process and the execution logic of the Alsys run-time environment. Since the Alsys scheduling algorithm involves proprietary information which was unavailable to us, reverse engineering experiments have been undertaken to deduce a plausibly defensible model for the Alsys scheduling algorithm. We are using controlled experimental probe Ada programs running in the Alsys environment. This document presents results pertaining to the elaboration phase of multi-package Ada programs. Complete reports [6, 14] on the work to date are available from the authors.

Bayesian inference for fault diagnosis in real-time distributed systems

We propose fault location strategies based on a Bayesian decision-theoretic approach. The proposed B-algorithm can locate the failed units in a distributed system with complexity which is linear in the number of subsystems in the network, making it suitable for hard real-time applications. The method is probabilistic and comparison-based, employing multiple incomplete test concepts.<<ETX>>

Bayes analysis for fault location in distributed systems

The authors propose a simple and practical probabilistic model, using multiple incomplete test concepts, for fault location in distributed systems using a Bayes analysis procedure. Since it is easier to compare test results among processing units, their model is comparison-based. This approach is realistic and complete in the sense that it does not assume conditions such as permanently faulty units, complete tests, and perfect or nonmalicious environments. It can handle, without any overhead, fault-free systems so that the test procedure can be used to monitor a functioning system. Given a system S with a specific test graph, the corresponding conditional distribution between the comparison test results (syndrome) and the fault patterns of S can be generated. To avoid the complex global Bayes estimation process, the authors develop a simple bitwise Bayes algorithm for fault location of S, which locates system failures with linear complexity, making it suitable for hard real-time systems. Hence, their approach is appealing both from the practical and theoretical points of view. >

The elaboration order problem of Ada

A serious problem in large Ada programs is finding a satisfactory order of elaboration for all of the programs compilation units. Some elaboration sequence must be selected but access to a program entity before its elaboration will raise the exception (PROGRAM_ERROR) shortly after execution of the program begins. The rules of the 1983 Ada standard1 fail to ensure that validated compilers produce a satisfactory elaboration order, or report an error if static analysis shows that such an order may not exist. The problem can make programs non‐portable and sensitive to modifications. This paper presents an algorithm ada_elab which determines an order of elaboration, free of PROGRAM_ERROR, if link‐time, static analysis can determine that there is one. Examples are used to illustrate the occurrence of elaboration errors and the inadequacy of pragma ELABORATE to resolve the problem. The paper closes with descriptions of elaboration and elaboration‐related problems that cannot be handled by ada_elab, access before initialization and coding practices that should avoid elaboration and initialization errors. While Ada is the subject of this paper, the problem being studied could occur in any language that permits a complex initialization of individual program modules prior to the execution of the main program module.

Formal model of uncertainty for possibilistic rules

Given a universe of discourse X--a domain of possible outcomes--an experiment may consist of selecting one of its elements, subject to the operation of chance, or of observing the elements, subject to imprecision. A priori uncertainty about the actual result of the experiment may be quantified, representing either the likelihood of the choice of x ∈ X or the degree to which any such x ∈ X would be suitable as a description of the outcome. The former case corresponds to a probability distribution, while the latter gives a possibility assignment on X.

Management and Scheduling of Tests for Locating Failures Dependent upon Operation-Time in Responsive Systems

We propose fault location strategies based on failure probabilities that are not constant over the system’s lifetime. The approach is not only to locate the failed units in a multiprocessor system but also to efficiently choose the appropriate diagnosis method. Point and set estimations are used for fault location inference. Methods are comparison based, employing multiple incomplete test concepts. A dynamic testing policy is given, which changes between single fault location and multiple fault location dependent upon system operation-time.