CN105335226A - Iterative static task list scheduling algorithm for multi-processor system - Google Patents

Abstract

The invention discloses an iterative static task list scheduling algorithm for a multi-processor system. The algorithm is characterized by being executed by the following steps: 1, setting an initial value of a current optimal task priority sequence, and taking a corresponding static list scheduling length as an initial value of a current optimal scheduling length; 2, obtaining a new task priority sequence from the current optimal task priority sequence, and if the corresponding static list scheduling length is less than the current optimal scheduling length, updating the current optimal scheduling length and the current optimal task priority sequence; 3, repeatedly executing the step 2 until the frequency of execution reaches a specified upper limit; 4, executing the steps 1 to 3 for each optimal task priority sequence; and 4, selecting a minimum value from all optimal scheduling lengths as a final scheduling result. According to the algorithm, the scheduling length is further reduced on the basis of a conventional static task list scheduling algorithm, so that the application execution efficiency is effectively improved.

Description

For the iterative static task list scheduling algorithm of multicomputer system

Technical field

The present invention relates to task scheduling field, specifically a kind of based on the iterative static task list scheduling algorithm on multicomputer system.

Background technology

In prior art, multicomputer system can meet the demand that application program performs Multi-task Concurrency to a greater extent, and the task scheduling how more effectively realized on multicomputer system still needs to be probed into further.At present, the list scheduling technology of Static task scheduling algorithm many employings task based access control graph model of main flow.In task graph model, represent application program with directed acyclic graph G=(V, E, W, C), a point represents a task, directed edge E between points _a,brepresent the follow-up restriction relation of forerunner between predecessor task a and subsequent tasks b, some weights W _arepresent the time of executing the task needed for a, directed edge E _a,bweight C _a,btime needed for expression task a communicates with b.Detailed introduction about task image model sees the research background declaratives of list of references " DynamicCriticalPathScheduling:AnEffectiveTechniqueforAll ocatingTaskGraphstoMultiprocessors ".Be generally model with task image due to Mission Scheduling, therefore, task scheduling also can be described as task image scheduling.

The implementation procedure of traditional static task static list dispatching algorithm is comparatively simple: according to task priority list, each task is scheduled in makes it obtain on the processing unit of earliest start time successively.Task priority list is obtained by special algorithm, and in priority list, the position at task place is more forward, and the priority that this task is scheduled is higher.About the detailed description of described static task static list dispatching algorithm see document " TaskSchedulingforParallelSystems " to the elaboration of Algorithm9 and Algorithm10.Static list dispatching algorithm only considers this kind of task priority sequence of task priority list, but task priority list may not be exactly OPTIMAL TASK priority sequence.Here, task priority sequence represents the precedence that task is scheduled.Therefore, only consider that a kind of static list dispatching algorithm of task priority sequence easily obtains poor scheduling result, namely larger scheduling length, thus cause application program execution efficiency lower.Described scheduling length represents the time of complete all required by task.

Summary of the invention

The present invention is for avoiding above-mentioned the deficiencies in the prior art, propose a kind of iterative static task list scheduling algorithm for multicomputer system, to on the basis of traditional static list dispatching algorithm, reduce scheduling length further, thus improve application program execution efficiency.

The present invention is that technical solution problem adopts following technical scheme:

A kind of iterative static task list scheduling algorithm for multicomputer system of the present invention, the feature that described multicomputer system performs N number of task on P processor is carried out in accordance with the following steps:

Step 1, definition outer circulation number of times are s; The threshold value of setting outer circulation number of times is S; Definition Inner eycle number of times is k; The threshold value of setting Inner eycle number of times is K;

Outer circulation number of times s=1 described in step 2, initialization;

Step 3, the task priority sequence of described task image is set at random under the s time outer circulation, thus obtains the OPTIMAL TASK priority sequence of described N number of task under the s time outer circulation, be designated as represent the OPTIMAL TASK priority sequence under described the s time outer circulation in m to be scheduled task; 1≤m≤N;

Step 4, with the OPTIMAL TASK priority sequence under described the s time outer circulation for task priority list, static task static list dispatching algorithm is utilized to dispatch task image, described in acquisition optimal scheduling length under the s time outer circulation, is designated as

Inner eycle number of times k=1 described in step 5, initialization;

Step 6, obtain two random number i and j from tandom number generator, and exchange the OPTIMAL TASK priority sequence under described the s time outer circulation in i-th task that is scheduled be scheduled task individual with jth thus the kth task priority sequence obtained under the s time outer circulation, be designated as represent the kth task priority sequence T under described s outer circulation ^{(s) (k)}in l to be scheduled task; 1≤i≤N, 1≤j≤N, 1≤l≤N;

Step 7, with the task priority sequence T of the kth under described the s time outer circulation ^{(s) (k)}for task priority list, utilize static task static list dispatching algorithm to dispatch task image, obtain the kth scheduling length under the s time outer circulation, be designated as SL ^{(s) (k)};

Step 8, to the kth scheduling length SL under described the s time outer circulation ^{(s) (k)}with the optimal scheduling length under the s time outer circulation compare; If optimal scheduling length then under described the s time outer circulation with OPTIMAL TASK priority sequence assignment is SL respectively ^{(s) (k)}and T ^{(s) (k)}; If optimal scheduling length under described the s time outer circulation with OPTIMAL TASK priority sequence all remain unchanged;

Step 9, by k+1 assignment to k, judge whether k≤K sets up; If set up, represent the optimal scheduling length under described the s time outer circulation not yet complete iteration, return step 6 and perform; Otherwise, perform step 10;

Step 10, by s+1 assignment to s, judge whether s≤S sets up; If set up, return step 3 and perform; Otherwise, represent S optimal scheduling length all complete iteration, and perform step 11;

Step 11, from described S optimal scheduling length in select the scheduling result of minimum value as described iterative static task list scheduling algorithm, be designated as SL _best, namely thus complete described iterative static task list scheduling algorithm, and with SL _bestweigh the performance height of described iterative static task list scheduling algorithm.

Compared with prior art, beneficial effect of the present invention is embodied in:

1, the present invention further considers multiple-task priority sequence on the basis of traditional static task static list dispatching algorithm, thus obtains less scheduling length, effectively improves application program execution efficiency.

2, the present invention adopts exchange in the priority of two Random Tasks generate T ^{(s) (k)}, this task priority generation strategy the most easily realizes.

3, the present invention in the s time outer circulation kth time Inner eycle with T ^{(s) (k)}for task priority list, static list dispatching algorithm is adopted to dispatch task image, by comparing SL ^{(s) (k)}with determine whether upgrading with this method makes monotone decreasing in an iterative process, thus ensure that the scheduling length that iterative static task list scheduling algorithm obtains must be not more than static task static list dispatching algorithm.

4, the present invention uses for reference the basic thought of particle cluster algorithm, take S as population scale, for particle status, S particle upgrades through K iteration completion status separately, optimum solution is chosen again, so just by adopting multiple-task priority sequence to avoid single task role priority sequence to be easily absorbed in the defect of locally optimal solution from the end-state of population;

5, the present invention can adjust the size of S and K according to the restriction of algorithm actual run time; Thus there is higher flexibility.

6, algorithm realization of the present invention clusters relative to employing task, the Static task scheduling algorithm of Task Duplication and dynamic priority list technique is more simple.

Accompanying drawing explanation

Fig. 1 is task image used in example of the present invention;

Fig. 2 is algorithm flowchart of the present invention.

Embodiment

In the present embodiment, on the multicomputer system that has four processors, the implementation of iterative static task list scheduling algorithm is roughly as follows:

1, set the initial value of current OPTIMAL TASK priority sequence, corresponding static list scheduling length is as the initial value of current optimal scheduling length;

2 generate new task priority sequence from current OPTIMAL TASK priority sequence, if the static list scheduling length of correspondence is less than current optimal scheduling length, then upgrade current optimal scheduling length and current OPTIMAL TASK priority sequence;

3 repeated execution of steps 2, until perform number of times to reach the appointment upper limit;

4 perform step 1 to 3 for each OPTIMAL TASK priority sequence;

5 select minimum value, as final scheduling result from all optimal scheduling length.

Specifically, as shown in Figure 2, dispatching the task image shown in Fig. 1, is carry out as follows:

Step 1, definition outer circulation number of times are s; Setting outer circulation threshold value is S; Contrast particle cluster algorithm, the scheduling length of a kind of task priority sequence and correspondence thereof can be regarded as the state of a particle in population, so S is the scale of this population; After S outer circulation, in population, the state of all particles all upgrades complete, from the end-state of all particles, select minimum scheduling length, and this scheduling length is final task scheduling result; In the present embodiment, S value is 4;

Definition Inner eycle number of times is k; Setting Inner eycle threshold value is K; K represents the upper limit of each particle state iterations; In the present embodiment, K value is 256; In general, S and K is larger, and the dispatching effect of described iterative Static task scheduling algorithm is better, but for the restriction of Riming time of algorithm, the value of S and K can not be too large;

Outer circulation number of times s=1 described in step 2, initialization;

Step 3, the task priority sequence of described task image is set at random under the s time outer circulation, thus obtains the OPTIMAL TASK priority sequence under the s time outer circulation of N number of task in described task image, be designated as represent the OPTIMAL TASK priority sequence under described the s time circulation in m to be scheduled task; 1≤m≤N;

For the task image shown in Fig. 1, a node corresponds to a task, as node V ₀corresponding to task 0; The weights of each node are tasks carrying and complete the required time, as described in the task 0 complete required time be 99; Article one, directed edge represents the follow-up restriction relation of forerunner on it between two tasks, as directed edge E _0,8on task 0 be the forerunner of task 8, task 8 is the follow-up of task 0, and namely the dispatching priority of task 8 can not higher than task 0; Weights on directed edge are relevant with call duration time required between two tasks on it, for example, if task 0 and 8 is scheduled in two different processing units respectively, then from task 0 to the time needed for the communication of task 8 be 55, if task 0 and 8 is all scheduled on same treatment unit, then from task 0 to the time needed for the communication of task 8 be 0;

Here task priority sequence represents the sequencing that in task image, each task is scheduled, and task priority sequence does not meet the follow-up constraint condition of forerunner and refers in task image that at least there is follow-up that a directed edge makes on it obtains higher priority than forerunner; For example, for the task image shown in Fig. 1, { 8,1,2,3,4,5,6,7,0,9,10,11,12,13,14} does not meet the follow-up constraint condition of forerunner to task priority sequence, because E _0,8on the priority of subsequent tasks 8 higher than predecessor task 0;

Here obtain the just initial value of s OPTIMAL TASK priority sequence, this value constantly monotone decreasing in subsequent step; For the task priority sequence not meeting the follow-up constraint of forerunner, forerunner can not be had precedence over be scheduled due to follow-up, thus can think that corresponding static list scheduling length is positive infinity; So, do not meet the task priority sequence of the follow-up constraint of forerunner and be not suitable as s OPTIMAL TASK priority sequence initial value;

In the present embodiment, in order to obtain the task priority sequence that four kinds meet the follow-up constraint of forerunner, four outer circulations adopt following four kinds of task priority sequence numerical procedures respectively: TL-Scheme, BL-Scheme, BL_TL-Scheme and Random-Scheme; Four to obtain the mode of task priority sequence different: TL-Scheme arranges each task according to TL weights ascending order, BL-Scheme is according to each task of BL weights descending sort, BL_TL-Scheme arranges each task according to status of a sovereign key word BL weights descending and secondary position key word TL weights ascending order, and Random-Scheme stochastic generation one meets the task priority sequence of forerunner's constraint.The computing method of task TL and BL weights are as shown in the formula shown in (1) and (2):

In the present embodiment, according to the computing method of described four kinds of task priority sequences, the initial value of each self-corresponding OPTIMAL TASK priority sequence of four outer circulations is as shown in table 1; The task image of contrast shown in Fig. 1, these four kinds of task priority sequences meet the follow-up constraint condition of forerunner really.

Table 1: four kinds of each self-corresponding task priority sequences of static task priority computing method

The present invention can adopt the task priority sequence computing method in paper " having the senior list dispatching method of communication contention Mission Scheduling in parallel embedded system " and " Listscheduling:extensionforcontentionawarenessandevaluat ionofnodeprioritiesforheterogeneousclusterarchitectures ", to ensure OPTIMAL TASK priority sequence initial value meets the follow-up constraint condition of forerunner;

Step 4, with described the s time circulation under OPTIMAL TASK priority sequence for task priority list, utilize static list dispatching algorithm to dispatch task image, obtain the optimal scheduling length under the s time circulation, be designated as

With certain task priority sequence for task priority list, static task static list dispatching algorithm is utilized to dispatch task image and the scheduling length obtained also can be called as the static list scheduling length that this task priority sequence pair is answered; Here, corresponding static list scheduling length the just initial value of s optimal scheduling length, to constantly be updated in an iterative process; In the present embodiment, with corresponding static list scheduling length with be respectively 824,731,731 and 829;

Inner eycle number of times k=1 described in step 5, initialization; K represents iterations;

Step 6, obtain two random number i and j from tandom number generator, then exchange the task priority sequence under described the s time circulation in i-th task that is scheduled be scheduled task individual with jth thus the kth exploratory task priority sequence obtained under the s time circulation; Be designated as represent the kth task priority sequence T under described the s time circulation ^{(s) (k)}in l to be scheduled task; 1≤i≤N, 1≤j≤N, 1≤l≤N;

In the present embodiment, as s=1 and k=1 time, from random number, two random numbers are obtained to generator: 1 and 4, exchange in the priority of task 1 and 4, obtain first task priority sequence T under first time outer circulation ^{(1) (1)}={ 0,4,2,3,1,5,6,7,8,10,9,11,13,12,14};

Step 7, with described the s time circulation under a kth task priority sequence T ^{(s) (k)}for task priority list, utilize static list dispatching algorithm to dispatch task image, obtain the kth scheduling length under the s time circulation, be designated as SL ^{(s) (k)};

In the present embodiment, T ^{(1) (1)}meet forerunner's constraint condition, with T ^{(1) (1)}for task priority list, obtain corresponding scheduling length SL according to static list dispatching algorithm ^{(1) (1)}=819, alternatively, T ^{(1) (1)}corresponding static list scheduling length SL ^{(1) (1)}=819; Illustrate in step 3, if T ^{(s) (k)}do not meet forerunner's constraint condition, then SL ^{(s) (k)}=+∞;

Step 8, to described the s time circulation under a kth scheduling length SL ^{(s) (k)}with the optimal scheduling length under the s time circulation compare; If optimal scheduling length under described the s time circulation with OPTIMAL TASK priority sequence assignment is SL respectively ^{(s) (k)}and T ^{(s) (k)}; If optimal scheduling length under described the s time circulation with OPTIMAL TASK priority sequence all remain unchanged;

This step makes monotone decreasing in the process of iteration, and then ensure that the scheduling length that iterative static task list scheduling algorithm obtains must be not more than static task static list dispatching algorithm; In the present embodiment, ${\mathrm{SL}}^{\left(1\right)\left(1\right)}=819<{\mathrm{SL}}_{best}^{\left(1\right)}=824,$ Thus renewal is made: ${\mathrm{SL}}_{best}^{\left(1\right)}={\mathrm{SL}}^{\left(1\right)\left(1\right)}$ And $T_{best}^{\left(1\right)}=T^{\left(1\right)\left(1\right)};$

Step 9, by k+1 assignment to k, judge whether k≤K sets up.If set up, represent the optimal scheduling length under described the s time outer circulation iteration also do not complete, return step 6 and perform; Otherwise, perform step 10;

Step 10, by s+1 assignment to s, judge whether s≤S sets up; If set up, then return step 3 and perform; Otherwise represent S optimal scheduling length all iteration completes, and continues to perform step 11;

Step 11, from described S optimal scheduling length in select the optimal scheduling length of minimum value as described iterative list scheduling algorithm, be designated as SL _best; Namely thus complete described iterative list scheduling algorithm; And with the optimal scheduling length SL of described iterative list scheduling algorithm _bestthe performance weighing described iterative list scheduling algorithm is good and bad; Scheduling length is all tasks carryings and completes required time loss, and thus the size of scheduling length weighs most important performance index of task scheduling algorithm performance;

In the present embodiment, four optimal scheduling length obtaining of described iterative static task list scheduling algorithm with be 690,690,595 and 622 respectively, and then obtain iterative static task list scheduling algorithmic dispatching result ${\mathrm{SL}}_{best}=min\{{\mathrm{SL}}_{best}^{\left(1\right)},{\mathrm{SL}}_{best}^{\left(2\right)},{\mathrm{SL}}_{best}^{\left(3\right)},{\mathrm{SL}}_{best}^{\left(4\right)}\}=595.$

Claims (1)

1., for an iterative static task list scheduling algorithm for multicomputer system, described multicomputer system performs N number of task on P processor, it is characterized in that carrying out in accordance with the following steps:

Step 1, definition outer circulation number of times are s; The threshold value of setting outer circulation number of times is S; Definition Inner eycle number of times is k; The threshold value of setting Inner eycle number of times is K;

Outer circulation number of times s=1 described in step 2, initialization;

Step 3, the task priority sequence of described task image is set at random under the s time outer circulation, thus obtains the OPTIMAL TASK priority sequence of described N number of task under the s time outer circulation, be designated as represent the OPTIMAL TASK priority sequence under described the s time outer circulation in m to be scheduled task; 1≤m≤N;

Step 4, with the OPTIMAL TASK priority sequence under described the s time outer circulation for task priority list, static task static list dispatching algorithm is utilized to dispatch task image, described in acquisition optimal scheduling length under the s time outer circulation, is designated as

Inner eycle number of times k=1 described in step 5, initialization;

Step 6, obtain two random number i and j from tandom number generator, and exchange the OPTIMAL TASK priority sequence under described the s time outer circulation in i-th task that is scheduled be scheduled task individual with jth thus the kth task priority sequence obtained under the s time outer circulation, be designated as represent the kth task priority sequence T under described s outer circulation ^{(s) (k)}in l to be scheduled task; 1≤i≤N, 1≤j≤N, 1≤l≤N;

Step 7, with the task priority sequence T of the kth under described the s time outer circulation ^{(s) (k)}for task priority list, utilize static task static list dispatching algorithm to dispatch task image, obtain the kth scheduling length under the s time outer circulation, be designated as SL ^{(s) (k)};

Step 8, to the kth scheduling length SL under described the s time outer circulation ^{(s) (k)}with the optimal scheduling length under the s time outer circulation compare; If optimal scheduling length then under described the s time outer circulation with OPTIMAL TASK priority sequence assignment is SL respectively ^{(s) (k)}and T ^{(s) (k)}; If optimal scheduling length under described the s time outer circulation with OPTIMAL TASK priority sequence all remain unchanged;

Step 9, by k+1 assignment to k, judge whether k≤K sets up; If set up, represent the optimal scheduling length under described the s time outer circulation not yet complete iteration, return step 6 and perform; Otherwise, perform step 10;

Step 10, by s+1 assignment to s, judge whether s≤S sets up; If set up, return step 3 and perform; Otherwise, represent S optimal scheduling length all complete iteration, and perform step 11;

Step 11, from described S optimal scheduling length in select the scheduling result of minimum value as described iterative static task list scheduling algorithm, be designated as SL _best, namely thus complete described iterative static task list scheduling algorithm, and with SL _bestweigh the performance height of described iterative static task list scheduling algorithm.