JPH05165397A - Scheduling device - Google Patents

Abstract

PURPOSE:To shorten the time required for a scheduling process by rearranging advance restriction relations between process tasks which should be considered in the scheduling process and generating necessary and sufficient advance restriction relations. CONSTITUTION:The scheduling process preprocess part 44 consists of an advance restriction input part 41, a dependency graph composition part 42, and a redundant restriction removal part 43. The redundant restriction removal part 43 detects and removes task-to-task restriction relations which can be substituted with a combination of other task-to-task restriction relations among advance restriction relations between plural process tasks by using a dependency graph which is generated by a dependency graph composition part 42 and shows the dependency relations between tasks in matrix expression. Therefore, the time required for the resource scheduling process in decentralized processes or parallel processes can be shortened.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a scheduling apparatus, and more particularly to a method for allocating a plurality of processing tasks having a preceding constraint relationship to a finite resource for processing requests in parallel processing and distributed processing.

[0002]

2. Description of the Related Art A scheduling device of this type is designed to satisfy the preceding constraint relationship of tasks and to obtain high processing efficiency when processing a plurality of processing unit tasks having preceding constraint relationships such as an order relationship with each other. First, the allocation of finite resources such as the processor, memory and peripheral devices required by the task is determined. This problem becomes a static scheduling problem when the precedence constraint relationships between all tasks and resource requirements are clear before task execution.

As a typical example of a conventional apparatus for solving this static scheduling problem, Japanese Patent Laid-Open No. 3-56983.
Some were published in the official gazette. FIG. 4 shows a flowchart showing a code generation process in the simulation code generation device for the plant simulation device. In the figure, the problem program is divided into tasks in ST1, and the pre-constraint relation between tasks, that is, the task causal relation is analyzed in ST2. Next, in ST3, it is determined from the analysis result of ST2 whether the scheduling method is static scheduling or dynamic scheduling. As a result, the task load measurement (S
T4), schedule determination (ST5), and execution control routine addition (ST6) are performed. On the other hand, if the dynamic scheduling is determined in ST3, a scheduling routine is added (ST7). Next, the processor allocation for the task is determined, and the parallel simulation code is output in ST8.

[0004]

Since the essence of the scheduling problem is a combinatorial problem, the complexity of the problem increases exponentially as the number of types of resources to be considered and the number of precedence constraint relationships increases. Therefore, the conventional scheduling device as described above has a problem that a large amount of time is required for the scheduling process.

The present invention has been made in order to solve such a problem. Scheduling is performed by organizing precedence constraint relationships between processing tasks to be considered in scheduling processing and generating necessary and sufficient precedence constraint relationships. It is an object of the present invention to obtain a scheduling device that can reduce the time required for processing.

[0006]

The scheduling device according to the invention of claim 1 of the present invention is capable of performing an alternative expression by a combination of a plurality of other task constraint relationships from among a plurality of preceding task constraint relationships. The present invention is characterized in that a redundant constraint removing means for detecting and removing a specific inter-task constraint relationship as a redundant inter-task constraint relationship is provided.

According to a second aspect of the present invention, dependency graph processing means for expressing the preceding constraint relationship of the entire processing task as a matrix expression G in the first dependency graph based on the resource access procedure of each processing task, and a plurality of other dependency graph processing means. From among the preceding constraint relationships among the processing tasks, the inter-task constraint relationships that can be represented by alternative combinations of other multiple task constraint relationships are detected as redundant task inter-task relationships and removed from the first dependency graph. It is characterized in that a redundant constraint removing means represented in the second dependency graph is provided.

According to the third aspect of the invention, the redundant constraint removing means calculates the m-reachable matrix H when m is 2 or more from the preceding constraint relationship G of the task represented by the matrix expression. Means and processing means for detecting and removing the inter-task constraint relationship appearing in both the matrix expression G and the m-reachability matrix H as a redundant inter-task constraint relationship.

[0009]

In the scheduling apparatus configured as described above, the redundant constraint removing means can be used to perform the alternative representation among the plurality of processing tasks among the preceding constraint relationships among the plurality of processing tasks by the combination of other constraint relationships among the tasks. Since the constraint relationship is detected and removed as a redundant inter-task constraint relationship, it is possible to reduce the inter-task constraint relationship to be processed during the actual scheduling process, and speed up.

Further, when the redundant constraint removing means is executed, the dependency graph processing means expresses it as the matrix expression G of the preceding constraint relationship of the entire processing task based on the resource access procedure of each processing task, so that it is processed by the matrix operation. Can be done, and the speed can be further increased.

[0011]

【Example】

Example 1. FIG. 1 shows the configuration of a scheduling device according to an embodiment of the present invention. In the figure, 41 is a preceding constraint input unit for inputting a preceding constraint relation, 42 is a dependency graph synthesizing unit for creating a dependency graph, 43 is a redundant constraint removing unit for detecting and eliminating redundant inter-task constraint relations, and 44 is scheduling. The preprocessing unit 45 is a scheduling unit.
The preceding constraint input unit 41 gives individual preceding constraint relationships, the dependency graph synthesizing unit 42 and the redundant constraint removing unit 43 synthesize necessary and sufficient task dependency graphs, and the scheduling unit 45 actually schedules and allocates resources to tasks. decide.

Next, the processing in the dependency graph synthesizing unit 42 will be described in more detail. The precedence constraint relationship of a processing task is defined by the execution procedure of the processing task and the access order of finite resources such as memory and input / output device. Therefore, the task precedence constraint relationship is analyzed, the access procedure of the processing task for each resource is checked, and is described in the resource access procedure table. Based on this table
Generate a first dependency graph. This first dependency graph is expressed by a matrix expression of (n × n) with the number of tasks n as the size. When the matrix element (i, j) is 1, it means that the task i has a constraint relationship that precedes j, and the matrix element (i, j)
When is 0, it means that there is no constraint relationship. To the matrix of the first dependency graph, the task precedence constraint relationship regarding each resource is written in order with reference to the resource access procedure table. However, the matrix elements for which the preceding constraint relationship between tasks has already been entered and whose value is 1 are left as 1. When the task precedence constraint relationships based on all resource access procedures are transferred to the matrix, the first dependency graph as shown in FIG. 2 is generated. In FIG. 2, the nodes of the first dependency graph have a processing task, and the paths connecting the nodes have an inter-task precedence constraint relationship. That is, N1 to N5 are nodes of the first dependency graph indicating processing tasks, and P1 to P7 are paths of the first dependency graph indicating constraint relationships between tasks. FIG. 2 and Expression 1 have the same meaning.

[0013]

[Equation 1]

Next, the processing in the redundant constraint removing section 43 will be described in more detail. Here, the redundant inter-task constraint relationship is one in which the order relationship of all tasks is the same even if there is no such constraint relationship and the scheduling results are equivalent. In the first dependency graph, reaching from one node to another node through m paths is called reaching in m steps. For example, in order to reach N1 to N4, it is reached in one step of P2 or 2 of P1 and P4.
Reachable in steps. A node set that can be reached in m-steps is called an m-reachable set, and a node set that can be reached within m steps is called an m-reachable set. Here, let m be a matrix representing the m-arrival set of each node of the first dependency graph.
-Defined as the reaching matrix.

Each element of the m-arrival matrix represents the number of m-arrival paths between tasks i and j. When the m-reachable sets are equivalent even if the connection relation of the first dependency graph is changed, m
-Reachability is preserved. A redundant path in the first dependency graph is a path whose context between tasks does not change even if the path does not exist.
One processing task can be reached via another route. That is, P2, P5 and P7 in FIG. 2 are redundant paths. The redundancy constraint removing unit 43 detects and removes such redundant paths and derives a graph with the minimum number of paths that preserves the reachability of the dependency graph.

First, with respect to the matrix expression G of the first dependency graph, a task state vector X having a length n representing the current task state is defined. The element x (k) of X is
It is 1 when the resource is allocated to the task k, and 0 when it is in the resource waiting state. Now, the task initial state X0
Since the reaching set with the number of steps m = 1 from is the node of all the destinations of the direct path from the resource allocation node, the matrix expression G of the first dependency graph and the m-arriving matrix are the same, and Can be obtained at. X1 = GX0 (2) Repeated in the same manner, the m-arrival set Xm from X0 is given by Equation 3
Can be obtained at. Xm = GXm-1 = G2 Xm-2 = ... = Gm X0 = Gm X0 ... (3) However, G1 = G.

Therefore, the m-reachable set Ym is expressed by equation (4). Ym = X1 + X2 + ... + Xm = (G1 + G2 + ... + Gm) Xm = FmXm ... (4) where (m> 1) -reachability matrix is defined by equation 5. H (m> 1) = G2 + G3 + ... + Gm ... (5)

The m-reachability matrix H (m> 1) has a node i, j if its element h (m> 1) (i, j) is not 0.
It means that the space can be reached in multiple steps. The redundant path of the graph is a direct path between nodes that can be reached in a plurality of steps, and is therefore a path of the dependency graph G that satisfies the condition of Expression 6. If h (m> 1) (i, j)> 0 and g (i, j) = 1, then g (i, j) = 0 (6) From Equation 6, the original first dependence graph G Thus, the second dependency graph G ′ with the minimum number of paths in which reachability is preserved is obtained by removing the redundant constraint from.

However, in the first dependency graph, when there is a loop route that returns to itself through some paths, the above-mentioned method determines that the paths forming the loop are redundant paths. Reachability is destroyed because the loop is broken. Therefore, this device cannot be applied when there is a loop in the first dependency graph.

The above is the principle of detection and removal of the redundant constraint relationship in the first embodiment. In this calculation, the storage area required for the calculation can be reduced by sequentially calculating Gm. This algorithm is shown below. 1. Gn = Gn-1 * G (G: first dependency graph) If Gn = 0, end 3. H (m> 1) = G2 + G3 + ... + Gn-1 4. If (h (m> 1) (i, j)> 0 and g (i,
If j)> 0), then g (i, j) = 0. The above 4 is performed for all i and j to obtain the necessary and sufficient second dependency graph G ′.

The redundant constraint is detected and removed from the first dependency graph shown in FIG. 2 according to the above algorithm. First, G2
Calculating each of ~ G4 results in Equation 7.

[0022]

[Equation 2]

From this, when H (m> 1) is calculated, equation 8 is obtained.
And the redundant path from P4 is P2.
(1,4), P5 (2,4), P7 (2,5).

[0024]

[Equation 3]

The necessary and sufficient second dependence graph G'is expressed by the equation (9).

[0026]

[Equation 4]

When this is expressed in a tree type, it becomes as shown in FIG. Since this embodiment presupposes a dependency graph without loops as the first dependency graph, the maximum reachable path length is at most the total number of tasks n. Therefore, always k +
Since Gk = 0 by 1 = n and the processing is completed, the stopping property of the processing is clear.

As is apparent from the above operation, the first embodiment
In the scheduling device according to, the time required for resource scheduling processing in distributed processing or parallel processing is shortened. In addition, the time and memory can be reduced as compared with the redundant path detection device using the backward search method.

It should be noted that noting that the value of the element of the above calculation is judged to be 0 or 1 or more, the matrix element is represented as a bit string, the product is calculated as a logical product, and the sum is calculated as a logical sum, A scheduling device that enables high-speed calculation and shortens the processing time can be obtained.

Further, in the first embodiment, the processing for deleting the redundancy constraint relationship has an independent structure as the scheduling pre-processing section 44, but only the redundancy constraint removing section 43 of the pre-processing section 44 is taken out and the dependency relationship is obtained. Even if it is realized as a part of the analysis device or a part of the scheduling device, the same effect as that of the first embodiment can be obtained.

[0031]

As described above, according to the first aspect of the present invention, in a scheduling apparatus that allocates a finite resource to a plurality of processing tasks having a preceding constraint relationship, among the preceding constraint relationships among a plurality of processing tasks, A scheduling device that can reduce the processing time can be obtained by providing a redundant constraint removing unit that detects and removes an inter-task constraint relationship that can be represented as an alternative by a combination of other multiple task constraint relationships as a redundant inter-task constraint relationship. It is effective.

Further, according to the invention of claim 2, in the scheduling device for allocating a finite resource to a plurality of processing tasks having a precedence constraint relation, the precedence constraint relation of the whole task is matrixed based on the resource access procedure of each treatment task. From among the dependency graph processing means represented in the first dependency graph as the expression G, and the preceding constraint relationships among the plurality of other processing tasks, an inter-task constraint relationship that can be represented by a combination of other plurality of inter-task constraint relationships can be represented. Is detected as a redundant inter-task constraint relationship from the first dependency graph and removed to remove the second
By providing the redundant constraint removing means represented in the dependency graph, there is an effect that a scheduling device that can further reduce the processing time can be obtained.

According to the invention of claim 3, claim 2
In the redundant constraint removing means in, the m-when m is 2 or more from the preceding constraint relation G of the task represented by the matrix expression.
By providing the calculating means for deriving the reachability matrix H and the processing means for detecting and removing the constraint relation between tasks appearing in both the matrix expression G and the m-reachable matrix H as a redundant constraint relation between tasks. Further, there is an effect that a scheduling device that can further reduce the processing time can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing a scheduling device according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a first dependency graph according to the first embodiment of the present invention.

FIG. 3 is an explanatory diagram showing a second dependency graph according to the first embodiment of the present invention.

FIG. 4 is a flowchart showing a processing procedure by a conventional scheduling device.

[Explanation of symbols]

 42 dependency graph synthesizing unit 43 redundancy constraint removing unit 45 scheduling unit

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] February 26, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 2

[Correction method] Change

[Correction content]

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0001

[Correction method] Change

[Correction content]

[0001]

BACKGROUND OF THE INVENTION This invention relates to scheduling apparatus, the processing request in particular parallel processing and distributed processing, relates to allocate the limited resources to the plurality of processing tasks with precedence constraints relationship.

[Procedure 3]

[Document name to be amended] Statement

[Correction target item name] 0007

[Correction method] Change

[Correction content]

According to a second aspect of the present invention, the dependency graph synthesizing means for expressing the preceding constraint relationship of the entire processing task as a matrix expression G in the first dependency graph based on the resource access procedure of each processing task, and other pluralities. From among the preceding constraint relationships among the processing tasks, the inter-task constraint relationships that can be represented by alternative combinations of other multiple task constraint relationships are detected as redundant task inter-task relationships and removed from the first dependency graph. It is characterized in that a redundant constraint removing means represented in the second dependency graph is provided.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0010

[Correction method] Change

[Correction content]

Further, when the redundant constraint removing means is executed, the dependency graph processing means expresses it as the matrix expression G of the preceding constraint relationship of the entire processing task based on the resource access procedure of each processing task, so that it is processed by the matrix operation. The redundant constraint can be removed at high speed .

[Procedure Amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0011

[Correction method] Change

[Correction content]

[0011]

EXAMPLES Example 1. FIG. 1 shows the configuration of a scheduling device according to an embodiment of the present invention. In the figure, 41 is a plurality of processes.
Prior constraint input section for inputting prior constraint relations of physical tasks , 4
2 is a dependency graph synthesizing means for creating a dependency graph , for example,
In this example, the dependency graph synthesizing unit inputs the preceding constraint input unit 41.
Access procedure of finite resources for each processing task
Based on, the matrix representation of the precedence constraint relationships for all processing tasks
Expression G is shown in the first dependency graph. Reference numeral 43 is a redundant constraint removing means for detecting redundant constraint relationships between tasks and removing them .
For example, here is the redundancy constraint removal unit,
Multiple other tasks from among the preceding constraint relationships among tasks
Constraints between tasks that can be represented as alternatives by combining constraint relationships
The first dependency graph, where the relationship is a redundant inter-task constraint relationship
From the second dependency graph. 44 is real
The scheduling pre-processing unit, which performs necessary processing before scheduling, performs the actual scheduling.
Cormorant is a scheduling unit. Before this scheduling
The processing unit 44 includes the preceding constraint input unit 41 and the dependency graph synthesis unit 4
2, the redundant constraint removing unit 43. Schematic location
The preceding constraint input unit 41 gives individual preceding constraint relationships, the dependency graph synthesizing unit 42 and the redundant constraint removing unit 43 synthesize necessary and sufficient task dependency graphs, and the scheduling unit 45 actually schedules and processes tasks. Determine resource allocations for.

[Procedure correction 6]

[Document name to be amended] Statement

[Correction target item name] 0012

[Correction method] Change

[Correction content]

Next, the processing in the dependency graph synthesizing unit 42 will be described in more detail. The precedence constraints related processing tasks each other, are defined and execution procedure of the processing task, the memory, the access order of the finite resources such as input and output devices. Therefore, the preceding constraint relationship between the processing tasks is analyzed, the access procedure of the processing tasks for each finite resource is checked, and the access procedure table is described. A first dependency graph is generated based on this table. This first dependency graph has the number of tasks n as its size (n × n)
Ru is represented by a matrix expression G. Matrix element (i, j) is 1
When it means that the task i has a constraint relation preceding j, and when the matrix element (i, j) is 0, it means that there is no constraint relation. For the matrix G of this first dependency graph,
With reference to the resource access procedure table, the task precedence constraint relationship for each resource is written in order. However, the matrix elements for which the preceding constraint relationship between tasks has already been entered and whose value is 1 are left as 1. When the task precedence constraint relationships based on all resource access procedures are transferred to the matrix, the first dependency graph as shown in FIG. 2 is generated. In FIG. 2, the nodes of the first dependency graph have a processing task, and the paths connecting the nodes have an inter-task precedence constraint relationship. That is, N1 to N
5 is a node of the first dependency graph indicating a processing task, P1 to P1
P7 is a path of the first dependency graph showing the constraint relationship between tasks. FIG. 2 and Expression 1 have the same meaning.

[Procedure Amendment 7]

[Document name to be amended] Statement

[Correction target item name] 0014

[Correction method] Change

[Correction content]

Next, the processing in the redundant constraint removing section 43 will be described in more detail. Here, the redundant inter-task constraint relationship is one of the precedence constraint relationships among multiple processing tasks.
Alternative table by combining multiple other task constraint relationships.
It refers to a constraint relationship between tasks that can be present. Even if there is no constraint relationship, the order relationship of all tasks is the same, and the scheduling results are equivalent. In the first dependency graph, reaching from one node to another node through m paths is called reaching in m steps. For example, in order to reach N4 from N1, it can be reached in one step of P2 or two steps of P1 and P4.
A node set that can be reached in m-steps is called an m-reachable set, and a node set that can be reached within m steps is called an m-reachable set. Here, m of each node of the first dependency graph
The matrix representing the reaching set is defined as the m-arriving matrix.

[Procedure Amendment 8]

[Document name to be amended] Statement

[Correction target item name] 0028

[Correction method] Change

[Correction content]

As is apparent from the above operation, the first embodiment
In the scheduling device according to, the redundancy constraint removing unit 43
Is included in the preceding constraint relationship between multiple processing tasks
Alternative table by combining multiple other task constraint relationships.
Representable inter-task constraint relationships as redundant inter-task constraint relationships
Then , the time required for the resource scheduling processing in the distributed processing or the parallel processing is shortened. In addition, the dependency graph synthesizing unit 42 processes each processing task.
Pre-constraint for the entire task based on these resource access procedures
The relationship is expressed as a matrix expression G in the first dependence graph, and
The long constraint removing unit 43 preempts other processing tasks.
A combination of multiple other constraint relationships between tasks
Redundant tasks with inter-task constraint relationships that can be expressed as alternatives
As an inter-constraint relation, it is detected and removed from the first dependency graph.
Since it is represented by the second dependency graph , the time and memory can be reduced as compared with the redundant path detecting device using the backward search method. Well
In particular, in the first embodiment, the redundancy constraint removing unit 43 uses the matrix table.
From the preceding constraint relation G of the task represented by the current formula, m is
Derivation of the m-reachability matrix H when it is 2 or more, and matrix expression G and
Constraint relation between tasks that appears in both m-reachability matrix H
Detecting and removing ties as redundant intertask constraint relationships
Has been realized in.

[Procedure Amendment 9]

[Document name to be amended] Statement

[Name of item to be corrected] 0032

[Correction method] Change

[Correction content]

Further, according to the invention of claim 2, in the scheduling device for allocating a finite resource to a plurality of processing tasks having a precedence constraint relation, the precedence constraint relation of the whole task is matrixed based on the resource access procedure of each treatment task. Dependency graph represented in the first dependency graph as expression G
From among the preceding constraint relationships among the synthesizing means and the plurality of other processing tasks, the inter-task constraint relationships that can be represented by a combination of the other plurality of inter-task constraint relationships are used as redundant inter-task constraint relationships as the first dependency. Second, detect and remove from graph
By having redundant constraint removal means depicted in dependency graph, the effect of scheduling apparatus is obtained can be shortened processing time.

[Procedure Amendment 10]

[Document name to be amended] Statement

[Correction target item name] 0033

[Correction method] Change

[Correction content]

According to the invention of claim 3, claim 2
In the redundant constraint removing means in, the m-when m is 2 or more from the preceding constraint relation G of the task represented by the matrix expression.
By providing the calculating means for deriving the reachability matrix H and the processing means for detecting and removing the constraint relation between tasks appearing in both the matrix expression G and the m-reachable matrix H as a redundant constraint relation between tasks. , the effect of scheduling apparatus is obtained can be shortened processing time.

Claims (3)

[Claims] 1. A scheduling device for allocating a finite resource to a plurality of processing tasks having a preceding constraint relationship,
Redundant constraint removing means for detecting and removing an inter-task constraint relationship that can be represented by a combination of other multiple inter-task constraint relationships as a redundant inter-task constraint relationship from the preceding constraint relationships among the plurality of processing tasks. A scheduling device characterized by being provided. 2. A scheduling device for allocating finite resources to a plurality of processing tasks having a precedence constraint relationship,
From among the preceding constraint relationships among the dependency graph processing means that represents the preceding constraint relationship of the entire task as the matrix expression G in the first dependency graph based on the resource access procedure of each of the processing tasks, and the preceding constraint relationships among the other processing tasks, A redundant constraint removing unit that detects a redundant inter-task constraint relationship that can be represented by an alternative combination of a plurality of other inter-task constraint relationships as a redundant inter-task constraint relationship, and removes it from the first dependency graph to represent the second dependency graph is provided. A scheduling device characterized by the above. 3. Redundant constraint removing means, a calculating means for deriving an m-reachable matrix H when m is 2 or more from the preceding constraint relation G of the task expressed by the matrix expression, and the matrix expression. 3. The scheduling device according to claim 2, further comprising processing means for detecting and removing the inter-task constraint relationship that appears in both G and the m-reachability matrix H as a redundant inter-task constraint relationship.