Monitoring Temporal Rules Combined with Time Series

Run-time monitoring of temporal properties and assertions is used for testing and as a component of execution-based model checking techniques. Traditional run-time monitoring however, is limited to observing sequences of pure Boolean propositions. This paper describes tools, which observe temporal properties over time series, namely, sequences of propositions with constraints on data value changes over time. Using such temporal logic with time series (LTLD) it is possible to monitor important properties such as stability, monotonicity, temporal average and sum values, and temporal min/max values. The paper describes the Temporal Rover and the DBRover, which are in-process and remote run-time monitoring tools, respectively, that support linear time temporal logic (LTL) with real-time (MTL) and time series (LTLD) constraints. 1. Temporal Logic and Run-time Monitoring Overview Temporal Logic is a special branch of modal logic that investigates the notion of time and order. In [6], Pnueli suggested using Linear-Time Propositional Temporal Logic (LTL) for reasoning about concurrent programs. Since then, several researchers have used LTL to state and prove correctness of concurrent programs, protocols, and hardware. Linear-Time Temporal Logic (LTL) is an extension of propositional logic where, in addition to the propositional logic operators there are four future-time operators and four dual past time operators: always in the future (always in the past), eventually, or sometime in the future (sometime in the past), until (Since), and next cycle (previous cycle). Metric Temporal Logic (MTL) was suggested by Chang, Pnueli, and Manna as a vehicle for the verification of real time systems [1]. MTL extends LTL by supporting the specification of relative time and real time constraints. All four LTL future time operators can be constrained by relative time and real time constraints specifying the duration of the temporal operator. This paper described additional extension to LTL and MTL suitable for the specification of time-series constraints. Run time Execution Monitoring (REM) is a class of methods of tracking temporal requirements for an underlying application. First applications of REM were verification oriented where REM methods were used to track whether an executing system conforms to formal specification requirements. Recent adaptations of REM methods enable run time monitoring for non-verification purposes such as temporal business rule checking and temporal security rule checking [5]. Unlike previously published methods [7], the newer methods are on-line, namely, temporal rules are evaluated without storing an ever growing and potentially unbounded history trace. The TemporalRover and DBRover tools described in this paper perform on-line REM using executable alternating finite automata. The technique enables on-line monitoring complex Kansas State Specification Pattern assertions at a rate of 6000 to 60,000 cycles per second on a 1GHz CPU [4], and is capable of monitoring past-time and future-time temporal logic augmented with real-time constraints, time-series constraints, and special counting operators described in [2]. High-speed on-line REM enables demanding applications such as formal specification based exception handling [3]. 2. Run Time Monitoring Tools: The Temporal Rover and DBRover The Temporal Rover [2] is a code generator whose input is a Java, C, C++, or HDL source code program, where LTL/MTL assertions are embedded as source code comments. The Temporal Rover parser converts this program file into a new file, which is identical to the original file except for the assertions that are now implemented in source code. The following example contains an embedded MTL assertion for a Traffic Light Controller (TLC) written using the Temporal Rover syntax asserting that for 100 milliseconds, whenever light is red, camera s.b. on: void tlc(int Color_Main, boolean CameraOn) { ... /* Traffic Light Controller functionality */ /* TRBegin TRClock{C1=getTimeInMillis()} // get time from OS TRAssert{ Always({Color_Main == RED} Implies Eventually_C1<1000_{CameraOn == 1}) } => // Customizable user actions {printf("SUCCESS");printf("FAIL");printf("DONE!");}