A Scheduling Analysis Method for Task Automation Based on Demand Upper Bound Function
By using the method based on the upper bound function of the requirements, the scheduling analysis of the time automaton is solved, the state space explosion problem is realized, and efficient analysis of larger-scale task models is achieved, and the calculation complexity and time are reduced.
Patent Information
- Application Number
- CN202111244760.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-26
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-10-26
AI Technical Summary
The prior art is prone to state space explosion problems when performing scheduling analysis of time automatons, resulting in high complexity and can only handle small-scale task models.
Using the task automata scheduling analysis method based on the demand upper bound function, it is transformed into the demand upper bound by establishing a task automata system model, generating a state migration system, establishing a linear planning model, dividing equivalent spaces, and using dynamic programming algorithm structures. Function problem, only the boundary state of the equivalent space is analyzed to reduce the computational complexity.
It effectively reduces the complexity and time of time automata scheduling analysis, can handle larger-scale task models, avoids the need to exhaust all state spaces, and improves analysis efficiency.
Smart Images

Figure CN113901671B_ABST
Abstract
Description
Technical field:
[0001] The invention belongs to the field of real-time scheduling and model detection of real-time systems, and in particular relates to a task automaton schedulability analysis method based on a demand upper bound function. Background technology:
[0002] Real-time systems are now more and more widely used in safety-critical systems such as automobiles and airplanes. These fields have extremely high safety requirements, so modeling and analysis of real-time systems are very important. Adjustability analysis is an important method for verifying the safety of real-time systems. The research on schedulability analysis of real-time systems mainly involves two areas: real-time scheduling and model checking, which are independent of each other. The real-time scheduling field often describes the model of the real-time system as an abstract and simple model, so the schedulability analysis of the model is often efficient and low in complexity, but such a model cannot be used to describe complex real-time systems; model checking aims to establish a suitable formal model based on the characteristics of the real-time system, and then perform schedulability analysis. Its basic idea is the exhaustive search of the state space. For the case where the state space is large, the problem of searching the state space explosion will occur. Therefore, for a model that can better characterize the real-time system, the more complex the schedulability algorithm of the model will be.
[0003] Timed automata are the most expressive models in the field of real-time systems. The graphical representation of timed automata is a directed graph, in which each node is associated with a task, and the clock constraints on the nodes and edges are used to constrain clock growth. Using traditional model detection methods to perform schedulability analysis on timed automata requires exhaustive search of the entire state space of the timed automata, which will cause the problem of state space explosion. To address this problem, existing state space simplification methods use the symbolic state of timed automata, such as the concepts of Region and Zone proposed by Alur et al., which can simplify the state space. However, these improvements still have complex analysis problems. For example, the schedulability analysis tool TIMES based on timed automata can only perform schedulability analysis on timed automata with 8 nodes and can only be used for small-scale task models. Summary of the invention:
[0004] In order to reduce the time of schedulability analysis of timed automata and solve the problem of search space explosion in the schedulability analysis process, the present invention proposes a task automata schedulability analysis method based on demand upper bound function, which specifically includes:
[0005] Step 1: Establish a task automation system model;
[0006] Establish a task automaton system τ = {T1, T2, T3…}. The user inputs the range of the number of timed automaton nodes, the maximum constant that appears in the constraints, and the range of the WCET and deadline of the task, automatically generates the timed automaton system, and adds tasks to the nodes of the timed automaton to form the task automaton system.
[0007] Step 2: Generate a corresponding state transition system for each task automaton of the task automaton system;
[0008] For each task automaton T of the task automaton system, a corresponding state transition system LTS(T) is generated. Each node (l, u) is called a state, l represents the node location in the automaton, u represents the set of clock values corresponding to the node, and the initial state is defined as (l0, u0), where u0 is initialized to {0, 0…}; LTS(T) generates a new state (l, u) from the initial state (l0, u0) through migration, and connects them to finally generate LTS(T).
[0009] Step 3: By establishing a linear programming model, solve the upper bound TBound of the number of dynamic programming iterations;
[0010] A linear programming model I is established to calculate the maximum utilization U(T) of the loop in LTS(T).
[0011] Establish linear programming model II to calculate the maximum load DEM that can be generated by the tasks released in the LTS(T) path max (T).
[0012] The boundary value TBound is calculated by formula (1), which is the upper limit of the number of calculations in the iterative calculation process.
[0013]
[0014] Step 4: Divide the state space of the task automaton into equivalent spaces RPzone.
[0015] Using formula (2), for any edge l in the task automaton i → j Create state space ξ ij , where I(l i ) represents node l i The invariant on , g represents the inequality constraint on the edge, and r represents the set of reset clocks.
[0016] ξ ij =g∩I(l i )∩[I(l j \r)] (2)
[0017] Repeat the above steps and put all the state spaces of the task automaton into set A.
[0018] All the ξ in the state space ij The state space is divided into RPzone, and the state space set to which any two states in the state space are the same after migration is the equivalent space RPzone; only the boundary states of the equivalent space are analyzed for schedulability, which solves the problem of state space explosion;
[0019] The state space ξ ij Divide into equivalent spaces, If any two states (l, u) and (l, u′) in the state space reach the same state space set after migration, the state space is called an equivalent space.
[0020] All state spaces in set A are repeatedly divided until all state spaces in set A are equivalent spaces.
[0021] Step 5: Use the dynamic programming algorithm structure to solve the demand upper bound function DBF(t), and transform the schedulability analysis problem of the task automaton into the problem of the demand upper bound function;
[0022] Using the concept of demand bound function (DBF) in the field of real-time scheduling, a schedulability analysis framework is established for the schedulability analysis of task automata. The schedulability analysis of task automata is transformed into calculating the task load released by the task automata within the time interval t, which is denoted as DBF. T (t).
[0023] Calculate DBF T (t) is equivalent to finding an optimal path with a length less than t in LTS(T) that releases the largest load of tasks. We use the four-tuple (DEM(π), DIS(l i ), LEN(π), (l i ,u i )) is used to abstractly represent the path π of LTS(T), recorded as a tuple. In the time interval t, DEM(π) is the sum of the loads of the tasks executed in the interval t, DIS(l i ) is the path from the first state (l1, u1) of the path π to the state (l i ,u i ) is the sum of the clock increments, LEN(π) is the maximum time required to complete the task within interval t, (l i ,u i ) is the final state of the path. The following (3)(4)(5) are DEM(π), DIS(l i), the calculation formula of LEN(π). IND(π) represents the set of tasks that are executed and completed in the path within the interval t, δ i Indicates the amount of clock increase in delayed migration.
[0024]
[0025]
[0026]
[0027] Create an initial tuple, using each state (l, u) in LTS(T) as a path, and use (6)(7) to initialize the tuple.
[0028] TUP1=(0,0,0,(l,u)) (6)
[0029] TUP2=(e(l),d(l),0,(l,u)) (7)
[0030] Determine whether the LEN of the tuple is less than t. If so, put the tuple into the set T.
[0031] Using the idea of dynamic programming, for each new quadruple TUP=(DEM, LEN, DIS, (l, u)) stored in the set T, the DEM, LEN and DIS of TUP are updated by formulas (8), (9), (10), and new quadruple TUP1 and TUP2 are obtained by formulas (11), (12).
[0032] DEM′=DEM+e(l′) (8)
[0033] DIS′=DIS+δ (9)
[0034] LEN′=max(LEN, DIS′+d(l′)) (10)
[0035] TUP1=(DEM′, LEN′, DIS′, (l′, u′)) (11)
[0036] TUP2=(DEM, LEN, DIS′, (l′, u′)) (12)
[0037] δ represents the minimum clock increase that (l,u) needs to go through to reach the state (l′,u′) to which it can migrate.
[0038] The use of equivalent space is reflected in the calculation of δ. The calculation process of δ is: by dividing the state space, we can get the set of equivalent state spaces that the state (l, u) can reach. Then δ1 is equal to (l, u) to The distance of the boundary state (using the boundary state to equivalently replace this state space); δ2 is equal to the distance from (l, u) to the boundary state. Update the quadruple according to different δ values according to formulas (8), (9), and (10).
[0039] If the LEN of the updated quadruple TUP1 is less than t and is better than all the quadruples corresponding to the state (l′, u’) in set T, then store TUP1 in set T, and the same applies to TUP2.
[0040] For any tuple newly stored in set T, new quadruples are generated by the method of tuple update until no new quadruples are stored in set T.
[0041] According to formula (13), calculate DBF corresponding to t from 1 to TBound T (t)
[0042] DBF T (t) = max{DEM|(DEM, LEN, DIS, (l, u) ∈ T ≤t}} (13)
[0043] Step 6: Calculate DBF(t) of the task automaton system, and perform schedulability determination according to the inequality DBF(t) < t.
[0044] According to formula (14), calculate DBF(t) of the task automaton system.
[0045]
[0046] If for DBF(t) < t, then the task automaton system is schedulable; otherwise, it is non - schedulable.
[0047] Furthermore, in the said step 3, the linear programming model I is to calculate the maximum utilization rate U(T) of the loop in the state transition graph; the formula for the utilization rate is U(T) = DEM(π) / DIS(π); in the linear programming model I, the loop length DIS(π) is normalized, and finding the maximum DEM(π) of the loop is the maximum utilization rate U(T) of the loop; the specific formula is:
[0048]
[0049]
[0050]
[0051] Furthermore, in step 3, the linear programming model II calculates the maximum load generated by the released tasks in the path in LTS(T); two auxiliary states are introduced into the model: (l src ,u src ) and, (l snk ,u snk ), add the auxiliary state task automaton T, record it as
[0052]
[0053]
[0054]
[0055]
[0056]
[0057]
[0058] Furthermore, in step 5, for the current ξ ij The reset constraints of any two clock variables xy in , three cases are considered in the process of dividing the equivalent space RPzone:
[0059] Current ij The clock variable xy does not reset. Use xy=c to partition the convex set ξ and find a minimum number c such that ξ ij ∩xy≤c is RPzone, ξ ij Divided into two parts
[0060]
[0061] If y is reset, use x=c to divide and find a minimum number c such that ξ ij ∩x≤c is RPzone, ξ ij Divided into two parts
[0062]
[0063] If x is reset, use y=c to divide and find a minimum number c such that ξ ij ∩y≤c is RPzone, ξ ij Divided into two parts
[0064]
[0065] Further, in the sixth and seventh steps, a quadruple (DEM(π), DIS(l i ), LEN(π), (l i , u i )) is used to abstractly represent the path π of the state transition system LTS(T). Through the idea of dynamic programming, the quadruple is updated, and only the quadruple that is optimal for (l i , u i ) is stored during the calculation process, reducing the computational complexity. Using DBF T (t) = max{DEM|(DEM, LEN, DIS, (l, u) ∈ T ≤t}} to calculate DBF T (t). For then the task automaton T is schedulable, otherwise, it is not schedulable.
[0066] Beneficial effects:
[0067] 1. The present invention equivalently partitions the state space. Through the equivalent partition, the set of state spaces reached by each state in the state space after migration is the same. In this way, only one boundary state can be used to equivalently replace all the states in this state space. Based on this new reduction method, exhaustive search of all state spaces is avoided, and only the boundary states are traversed for schedulability analysis, reducing the complexity of schedulability analysis;
[0068] 2. The present invention utilizes the DBF (Demand Bound Function) framework in the real-time domain, that is, the demand upper bound function framework. Briefly speaking, it is necessary to calculate the value of DBF(t) corresponding to each time t, and judge schedulability according to the inequality. If DBF(t) < t, it is schedulable, otherwise it is not schedulable. Implementation process: First, a boundary value Tbound of time t needs to be calculated. Only in the interval [1, Tbound] may there be a non-schedulable situation (this is the property of this framework, so it is only necessary to verify DBF(t) for each t in the interval [1, Tbound]). For each t in [1, Tbound], calculate DBF(t). If the inequality DBF(t) < t is satisfied for all of them, then it is schedulable, otherwise it is not schedulable. And the dynamic programming algorithm structure is used to solve DBF(t) for schedulability analysis of the states in the state space. For each t in [1, Tbound], calculate DBF(t). First, find DBF(1), and then find DBF(2) based on DBF(1), and so on. Based on the value of DBF(t - 1), find DBF(t).
[0069] 3. The present invention uses a linear programming model to solve the upper bound of the number of iterations during the dynamic programming calculation process, reducing the computational complexity. Description of the drawings:
[0070] Figure 1 is a flow chart of the steps of the method of the present invention;
[0071] Figure 2 Schematic diagram of the task automaton.
[0072] Figure 3 for Figure 2 The state transition system LTS(T) diagram of the task automaton shown;
[0073] Figure 4 Schematic diagram of the equivalent space partitioning situation without clock reset.
[0074] Figure 5 Schematic diagram of the equivalent space division of y clock reset.
[0075] Figure 6 Schematic diagram of the equivalent space partitioning for x clock reset.
[0076] Figure 7 Schematic diagram for finding the distance from clock u to the equivalent space.
[0077] Figure 8 This is an experimental diagram with U as the variable.
[0078] Fig. 9 This is an experimental diagram with the number of task automata as a variable.
[0079] Fig.10 This is an experimental diagram with the maximum constant as the variable. Specific implementation method:
[0080] The following embodiments are described by specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The top details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the features in the following embodiments and implementations can be combined with each other without conflict. It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention.
[0081] The present invention provides a task automation schedulability analysis method based on demand upper bound function, such as Figure 1 As shown:
[0082] Step 1: Establish a task automation system model.
[0083] Establish a timed automaton system τ = {T1, T2, T3...}. The user inputs the range of the number of timed automaton nodes, the maximum constant appearing in the constraints, and the range of the WCET and deadline of the task, automatically generates a timed automaton system, and adds tasks to the nodes of the timed automaton to form a task automaton system.
[0084] The timed automaton models used in this embodiment are randomly generated according to the code based on the relevant parameters input by the user. The scale parameters of the timed automaton input by the user are shown in Table 1, the parameters of adding tasks to the nodes of the timed automaton are shown in Table 2, and the task automaton is shown in Table 3. Figure 2 .
[0085] Table 1 Timed automaton system parameters
[0086]
[0087]
[0088] Table 2 Task parameters
[0089]
[0090] Step 2: Generate a corresponding state transition system for each task automaton of the task automaton system;
[0091] For each task automaton T of the task automaton system, a corresponding state transition system LTS(T) is generated. Figure 3 Each node (l, u) is called a state, l represents the node location in the automaton, u corresponds to the set of clock values, and the initial state is defined as (l0, u0), where u0 is initialized to {0, 0…}; LTS(T) migrates from the initial state (l0, u0) to generate a new state (l, u), and connects them to finally generate LTS(T).
[0092] LTS(T) is expressed as a directed graph. Figure 3 Each node in is called a state, and each state is a reachable state that can be migrated from the initial state of the task automaton. Each state is represented in the form of (l, u), where l is the node location in the task automaton, and u represents the set of corresponding clock values. The initial state is defined as (l0, u0), and each clock value in the initial state is 0; Figure 3 Each edge in corresponds to a migration in the generation process. Figure 3 There are two migration methods:
[0093] Delay transition: If the clock value set u' obtained after the clock synchronization of state (l, u) still satisfies the invariant condition on node l, then state (l, u) generates a new state (l, u') pointed by an edge. The two states before and after delayed transition are on the same node, but the clock value increases and the node does not change.
[0094] Discrete transition: When the clock value u in state (l, u) satisfies the inequality condition on the edge connecting to the adjacent node l', and the clock value is updated according to the clock reset condition to obtain u', if u' satisfies the invariant condition on l', then the state (l, u) generates a new state (l', u') pointed to by an edge. The two states before and after the discrete transition are located at different locations. Figure 2 The corresponding state transition system is as follows Figure 3 .
[0095] Step 3: By establishing a linear programming model, solve the upper bound Tbound of the number of dynamic programming iterations;
[0096] (1) Establish a linear programming model I to calculate the state transition system (such as Figure 3 ) is the maximum utilization rate U(T) of the loop in the state transition diagram. The calculation formula of the utilization rate is U(T) = DEM(π) / DIS(π). DEM(π) is the load of the task released by loop π in the state transition diagram, and DIS(π) represents the length of the loop. In model I, the loop length is normalized to calculate the maximum utilization rate U(T) of the loop.
[0097] In model I, η represents any edge (l, u)→(l', u') in LTS(T), and X(η) represents a variable indicating whether the path to be sought passes through edge η in LTS(T), with a value range of [0, 1]. If the path to be sought passes through edge η, then X(η)>0; if it does not pass through edge η, then X(η)=0. For discrete migration (l, u)→(l', u'), e(η)=e(l); for delayed migration (l, u)→(l, u'), Δ(η) is equal to the clock growth amount. Ω + (l,u) represents the set of edges flowing into state (l,u), Ω - (l,u) represents the set of edges flowing out of state (l,u), Ω a (T) represents the edge where discrete migration occurs, Ω ↑ (T) represents the edge where delayed migration occurs, Ω ↑ Indicates delayed migration.
[0098]
[0099]
[0100]
[0101] (2) Establish linear programming model II to calculate the maximum load DEM generated by the release task in the path in LTS(T) max (T). Model II is as follows, introducing two auxiliary states into the model: (lsrc, u src ) and, (l snk ,u snk ), we add the auxiliary state task automaton T as Add an edge (l to any state (l,u) of LTS(T) src ,u src )→(l,u) and (l,u)→(l src ,u src ), remember to do η represents The edge (l, u) → (l', u') in the graph, Y(η) indicates whether the required path passes through The variable of edge η in , takes the value of 0 or 1. If the path in question passes through edge η, then Y(η) = 0, if not, then Y(η) = 1. e(η) = e(l'). Model II is as follows.
[0102]
[0103]
[0104]
[0105]
[0106]
[0107]
[0108] The bound value Tbound is calculated by formula (1), which is the upper bound of the number of calculations in the iterative calculation process.
[0109]
[0110] Step 4: Divide the state space of the task automaton into equivalent spaces RPzone;
[0111] Using formula (2), for any edge l in the task automaton i → j Create state space ξ ij , where I(l i ) represents the invariant on node i, g represents the inequality constraint on the edge, and r represents the set of reset clocks.
[0112] ξ ij =g∩I(l i )∩[I(l j \r)] (2)
[0113] Repeat the above steps and put all the state spaces of the task automaton into set A.
[0114] All the ξ in the state space ij To divide, If any two states (l, u) and (l, u') in the state space reach the same state space set after migration, then the state space is called the equivalent space RPzone, which enables us to perform schedulability analysis only on the boundary states of the equivalent space. This state space simplification method solves the problem of state space explosion.
[0115] For the reset constraints of any two clock variables xy in the currently divided state space, three cases are considered in the process of dividing the equivalent space RPzone:
[0116] 1. Current ij The clock variable xy does not reset, and xy=c is used to partition the convex set ξ ij , find a minimum number c such that ξ ij ∩xy≤c is RPzone, such as Figure 4 ,ξ ij Divided into two parts
[0117]
[0118] 2. If y is reset, use x=c to divide and find a minimum number c such that ξ ij ∩x≤c is RPzone, such as Figure 5 ,ξ ij Divided into two parts
[0119]
[0120] 3. If x is reset, use y=c to divide and find a minimum number c such that ξ ij ∩y≤c is the RPzone, such as Figure 6 ,ξ ij Divided into two parts
[0121]
[0122] Subdivide all state spaces in set A until all state spaces in A are equivalent spaces RPzone.
[0123] Step 5: Use the dynamic programming algorithm structure to solve the demand upper bound function DBF(t), and transform the schedulability analysis problem of the task automaton into the problem of the demand upper bound function;
[0124] Using the concept of demand bound function (DBF) in the field of real-time scheduling, a schedulability analysis framework is established for the schedulability analysis of task automata. The schedulability analysis of task automata is transformed into calculating the sum of the maximum loads generated by the task automata releasing tasks within a time interval t, which is recorded as DBF. T (t).
[0125] Calculate DBF T (t) is equivalent to finding an optimal path with a length less than t in LTS(T) that releases the largest load of tasks. We use the four-tuple (DEM(π), DIS(l i ), LEN(π), (l i ,u i )) is used to abstractly represent the path π of LTS(T), recorded as a tuple. In the time interval t, DEM(π) is the load of the tasks completed in the interval t; DIS(l i ) is the path from the first state (l1, u1) of the path π to the state (l i ,u i ) is the sum of the clock increments; LEN(π) is the maximum time required to complete a task within interval t; (l i ,u i ) is the final state of the path. The following (3)(4)(5) are DEM(π), DIS(l i ), the calculation formula of LEN(π).
[0126]
[0127]
[0128]
[0129] IND(π) represents the set of tasks that have been executed and completed in the path. For any released task, only the task whose release time and completion time are within the time interval t can be put into IND(π). i Indicates the amount of clock increase in delayed migration.
[0130] Create an initial tuple, using each state (l, u) in LTS(T) as a path, and use formulas (6) and (7) to initialize the tuple.
[0131] TUP1=(0,0,0,(l,u)) (6)
[0132] TUP2=(e(l),d(l),0,(l,u)) (7)
[0133] Determine whether the LEN of the tuple is less than t. If so, put the tuple into the set T.
[0134] Using the idea of dynamic programming, for each new quadruple TUP=(DEM, LEN, DIS, (l, u)) stored in the set T, the DEM, LEN and DIS of TUP are updated by formulas (8), (9), (10), and new quadruple TUP1 and TUP2 are obtained by formulas (11), (12).
[0135] DEM′=DEM+e(l′) (8)
[0136] DIS′=DIS+δ (9)
[0137] LEN′=max(LEN, DIS′+d(l′)) (10)
[0138] TUP1=(DEM′,LEN′,DIS′,(l′,u′))(11)
[0139] TUP2=(DEM, LEN, DIS′, (l′, u′)) (12)
[0140] δ represents the minimum clock growth that (l, u) needs to experience to reach the state (l′, u′) to which it can migrate, such as Figure 7 .
[0141] If the LEN of the updated quadruple TUP1 is less than t and is better than all quadruple corresponding to the state (l′, u′) in the set T, TUP1 is stored in the set T, and the same applies to TUP2.
[0142] For any new tuple stored in set T, a new quadruple is generated by the tuple update method until no new quadruple is stored in set T. DBF is calculated according to formula (13): T (t).
[0143] According to formula (13), calculate the DBF corresponding to t from 1 to TBound T (t)
[0144] DBF TDBF(t) = max{DEM | (DEM, LEN, DIS, (l, u)) ∈ T ≤t} (13)
[0145] The specific process can be summarized as follows: Initialize the quadruple, update the quadruple according to the formula, only store the quadruples with LEN < 1, and the largest DEM in these quadruples is DBF(1); update the quadruples stored in 1, only store the quadruples with the updated LEN < 2, and the largest DEM in these quadruples is DBF(2), and so on.
[0146] Step 6: Calculate DBF(t) of the task automaton system, and perform schedulability determination according to the inequality DBF(t) < t.
[0147] According to formula (14), calculate DBF(T) of the task automaton system from t = 1 to TBound.
[0148]
[0149] If for then the automaton is schedulable; otherwise, it is non - schedulable, and output the schedulability of the task automaton.
[0150] The experimental results prove that the partitioning can greatly reduce the schedulability analysis time. The experimental results are as Figure 8 、 9 、10. The vertical axis represents the schedulability analysis time, the horizontal axis represents the variables, N represents the number of task automata in each task automaton system, U represents the utilization rate of the task automaton system, and Ψ represents the largest constant appearing in the task automaton constraints. t pr represents the analysis time in the case of partitioning, and t np represents the analysis time in the case without partitioning.
[0151] Obviously, the above - mentioned embodiments of the present invention are merely examples for clearly explaining the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.
Claims
1. A task automaton schedulability analysis method based on demand upper bound function, applied to automobile and aircraft safety systems, characterized by: Step 1: Establish a task automation system model; Step 2: Generate a corresponding state transition system for each task automaton of the task automaton system; Step 3: By establishing linear programming model I and linear programming model II, solve the upper bound Tbound of the number of dynamic programming iterations; Step 4: Divide the state space of the task automaton into equivalent spaces RPzone; Step 5: Use the dynamic programming algorithm structure to solve the demand upper bound function DBF(t), and transform the schedulability analysis problem of the task automaton into the problem of the demand upper bound function; Step 6: Calculate DBF(t) of the task automaton system and perform schedulability analysis; In the step 3, the linear programming model I is used to calculate the maximum utilization rate U(T) of the loop in the state transition diagram; The calculation formula of utilization is U(T)=DEM(π) / DIS(π); in the linear programming model I, the loop length DIS(π) is normalized, and the maximum loop DEM(π) is the maximum loop utilization U(T); Ω ↑ Indicates delayed migration. The specific formula is: In step 3, the linear programming model II calculates the maximum load generated by the released tasks in the path in LTS(T); two auxiliary states are introduced into the model: (l src ,u src ) and,(l snk ,u snk ), add the auxiliary state task automaton T, record it as 2. The method for analyzing the schedulability of task automata based on the demand upper bound function according to claim 1 is characterized by: In the step 5, for the current ξ ij The reset constraints of any two clock variables xy in , three cases are considered in the process of dividing the equivalent space RPzone: Current ij The clock variable xy does not reset. Use xy=c to partition the convex set ξ and find a minimum number c such that ξ ij ∩xy≤c is RPzone, ξ ij Divided into two parts If y is reset, use x=c to divide and find a minimum number c such that ξ ij ∩x≤c is RPzone, ξ ij Divided into two parts If x is reset, use y=c to divide and find a minimum number c such that ξ ij ∩y≤c is RPzone, ξ ij Divided into two parts 3. The method for analyzing the schedulability of task automata based on demand upper bound function according to claim 1 is characterized in that: In step 6, the four-tuple (DEM (π), DIS (l i ),LEN(π),(l i ,u i )) is used to abstractly represent the path π of the state transition system LTS(T). Through the idea of dynamic programming, the quaternary is updated, and only the path for (l i ,u i ) The optimal quadruple, reducing computational complexity, using DBF T (t)=max{DEM|(DEM,LEN,DIS,(l,u)∈T ≤t }Calculate DBF T (t); for Then the task automaton T is schedulable, otherwise it is unschedulable.
