CN109739638A - A kind of EDF schedulability determination method and device based on deep learning - Google Patents
A kind of EDF schedulability determination method and device based on deep learning Download PDFInfo
- Publication number
- CN109739638A CN109739638A CN201811486770.6A CN201811486770A CN109739638A CN 109739638 A CN109739638 A CN 109739638A CN 201811486770 A CN201811486770 A CN 201811486770A CN 109739638 A CN109739638 A CN 109739638A
- Authority
- CN
- China
- Prior art keywords
- task
- schedulability
- pending
- subset
- edf
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The present invention provides a kind of EDF schedulability determination method and device based on deep learning, this method comprises: first, obtain pending task-set, then, pending task-set is divided using space-division method, obtains first task subset, according to the dispatch feature of first task subset and preparatory trained schedulability decision model, schedulability is obtained to determine as a result, schedulability determines result for indicating that can first task subset be scheduled in single core processor according to EDF algorithm.Can method provided by the present invention carry out schedulability judgement, improve the accuracy that schedulability determines result using the schedulability decision model that can be predicted task-set and be scheduled in single core processor according to EDF algorithm trained in advance.In addition, carrying out schedulability judgement to the corresponding first task subset of pending task-set, additionally it is possible to when complete pending task-set can not be scheduled, provide decision-making foundation.
Description
Technical field
The present invention relates to Real-Time Scheduling field more particularly to a kind of EDF schedulability determination methods based on deep learning
And device.
Background technique
Real-time system is mainly towards the application system being closely related with time factor, it is in Aeronautics and Astronautics, medical treatment, silver
The multiple fields such as row, multimedia suffer from relatively broad application.Since real-time system has strict demand to time behavioral trait,
Therefore, in the design of real-time system, the predictability of real-time system is designer's key factor in need of consideration.Wherein, real
When system predictability usually it is related with dispatching algorithm employed in real-time system.
For periodic task, dynamic dispatching is frequently with earliest Deadline First (Earliest priority-based
Deadline First, referred to as: EDF) algorithm.EDF algorithm specific manifestation are as follows: formula is determined by schedulability, obtains task
The schedulability of collection determine as a result, and determined according to schedulability as a result, according to task deadline dynamic dispatching task-set
In each task.
But when carrying out dynamic dispatching using EDF algorithm, need to pass through under the premise of a large amount of hypothesis and constraint condition
Mathematical method, which is derived, can determine that formula, this schedulability caused determine with the schedulability of Formal Language Description
Result precision is lower, and the schedulability determines that result can only determine that can complete task collection be scheduled, and leads to dynamic dispatching
Flexibility it is lower.
Summary of the invention
The present invention provides a kind of EDF schedulability determination method and device based on deep learning, to realize that raising is adjustable
Spend the flexibility of sex determination result precision and dynamic dispatching.
In a first aspect, the present invention provides a kind of EDF schedulability determination method based on deep learning, this method comprises:
Obtain pending task-set, wherein the pending task-set includes at least one periodically and preemptible waits for
Execution task；
The pending task-set is divided using space-division method, obtains first task subset, wherein described
First task subset includes at least one described pending task, and the space-division method is used for by the pending task
The high-dimensional space of property parameters mapped divided, to obtain the different arrangement groups of the pending task spatially
It closes；
According to the dispatch feature of the first task subset and preparatory trained schedulability decision model, obtain adjustable
Spend sex determination result, wherein the schedulability determines result for indicating that can the first task subset be handled in monokaryon
It is scheduled according to earliest Deadline First EDF algorithm on device.
Further, the dispatch feature and trained schedulability judgement in advance according to the first task subset
Model, acquisition schedulability determine after result, further includes:
According to schedulability judgement as a result, can be scheduled in single core processor according to EDF algorithm, and it is pending
The most first task subset of task quantity is determined as goal task subset, so that single core processor is according to EDF algorithm tune
Spend each pending task in the goal task subset.
Further, the pending task-set includes the M pending tasks, and M is the integer greater than 0；
It is described that the pending task-set is divided using space-division method, obtain first task subset, comprising:
Fully intermeshing is carried out to the M pending tasks, is obtainedA first task subset,
Wherein,Indicate that the I progress obtained first task subset of permutation and combination is chosen from the M pending tasks
Number, I are the integer greater than 0, and X is the number that the M pending tasks are carried out with the first task subset that fully intermeshing obtains, X
For the integer greater than 1.
Further, the trained schedulability decision model in advance obtains in the following manner:
Obtain task-set, wherein the task-set includes at least one task；
The task-set is divided using space-division method, obtains the second task subset, wherein described second
Business subset carries schedulability and determines that label, the schedulability determine label for indicating that can the second task subset
Scheduled according to EDF algorithm in single core processor, the second task subset includes at least one described task；
Label is determined according to the dispatch feature of the second task subset and the schedulability, to deep neural network
DNN model is trained, and obtains the schedulability decision model.
Further, the dispatch feature is extracted in the following manner:
According to behavioural information and attribute information of the task when scheduled according to EDF algorithm in single core processor, building the
One dispatch feature set, wherein the first dispatch feature set includes at least one first dispatch feature；
According to preset condition, selected at least one described first dispatch feature that the first dispatch feature set includes
Select the second dispatch feature；
Data processing is carried out to second dispatch feature, obtains the dispatch feature.
Further, the property parameters include following one or more:
Duty cycle, opposite deadline and the worst execution time.
Further, the dispatch feature includes following one or more:
Central processor CPU utilization rate, density, scheduled complexity, cutoff rate, implementation rate, scheduling rate, effectively scheduling
Rate is preempted rate, is effectively preempted rate, idleness, effective idleness, superperiod load factor, synchronizes busy cyclic loading rate, is special
Period load factor.
Second aspect, the present invention also provides a kind of EDF schedulability decision maker based on deep learning, the device packet
It includes:
Module is obtained, for obtaining pending task-set, wherein the pending task-set includes that at least one is pending
Task；
Division module obtains first task for dividing using space-division method to the pending task-set
Subset, wherein the first task subset includes at least one described pending task, the space-division method indicate to by
The high-dimensional space of property parameters mapped of the pending task is divided, and obtains the pending task spatially
Different permutation and combination；
Determination module, for the dispatch feature and trained schedulability judgement in advance according to the first task subset
Model obtains schedulability and determines result, wherein the schedulability determines that result indicates that can the first task subset
It is scheduled according to EDF algorithm in single core processor.
The third aspect the present invention also provides a kind of EDF schedulability decision maker based on deep learning, the device include:
Memory and processor；
The memory stores program instruction；
The processor executes described program instruction, to execute method described in first aspect.
Fourth aspect, the present invention also provides a kind of storage mediums, comprising: program；
Described program is when being executed by processor, to execute method described in first aspect.
The present invention provides a kind of EDF schedulability determination method based on deep learning, this method comprises: obtaining pending
Task-set, pending task-set includes at least one periodical and pending task of preemptible, using space-division method pair
Pending task-set is divided, and obtains first task subset, wherein first task subset includes at least one pending
Business, space-division method to by the high-dimensional space of property parameters mapped of pending task for dividing, to obtain
The different permutation and combination of pending task spatially, further, according to the dispatch feature of first task subset and training in advance
Good schedulability decision model obtains schedulability and determines as a result, schedulability determines result for indicating first task
Can collection scheduled according to earliest Deadline First EDF algorithm in single core processor.
The beneficial effect of method provided by the present invention is:
Using in advance it is trained can predict task-set can in single core processor according to EDF algorithm be scheduled can
Sex determination model is dispatched, schedulability judgement is carried out, improves the accuracy that schedulability determines result.In addition, to pending
The corresponding first task subset of task-set carries out schedulability judgement, additionally it is possible to can not be scheduled in complete pending task-set
When, decision-making foundation is provided.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without any creative labor, can be with
It obtains other drawings based on these drawings.
Fig. 1 is that the process of the EDF schedulability determination method embodiment one provided by the invention based on deep learning is illustrated
Figure；
Fig. 2 is that the process of the EDF schedulability determination method embodiment two provided by the invention based on deep learning is illustrated
Figure；
Fig. 3 is that the process of the EDF schedulability determination method embodiment three provided by the invention based on deep learning is illustrated
Figure；
Fig. 4 is the structural schematic diagram that the EDF schedulability decision maker provided by the invention based on deep learning implements one；
Fig. 5 is the structural schematic diagram that the EDF schedulability decision maker provided by the invention based on deep learning implements two；
Fig. 6 is the structural schematic diagram that the EDF schedulability decision maker provided by the invention based on deep learning implements three；
Fig. 7 is the structural representation of the EDF schedulability decision maker example IV provided by the invention based on deep learning
Figure.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Fig. 1 is that the process of the EDF schedulability determination method embodiment one provided by the invention based on deep learning is illustrated
Figure.EDF schedulability determination method provided in this embodiment based on deep learning can be schedulable by the EDF based on deep learning
Sex determination device (hereinafter referred to as are as follows: schedulability decision maker) executes, which can be the equipment such as processor, computer,
The present invention is with no restrictions.
As shown in Figure 1, the method for the present embodiment the following steps are included:
S101, pending task-set is obtained.
Wherein, pending task-set includes at least one periodical and pending task of preemptible, i.e., pending task
With the periodical and property seized.
Schedulability decision maker obtains pending periodical and the property seized present in current real-time system
Business.Schedulability decision maker periodically can send acquisition instruction to real-time system, exist to obtain in current real-time system
Pending task, when can also occur pending task in real-time system, actively pending task sent by real-time system
To schedulability decision maker.The quantity for the pending task that schedulability decision maker obtains can be one, be also possible to
It is multiple.When the quantity of pending task is one, correspondingly, pending task-set includes unique pending task.When to
When the quantity of execution task is multiple, correspondingly, pending task-set includes multiple pending tasks.
It should be noted that general use is converted into week for the acyclic task occurred in real-time system
The method of phase property task is scheduled, for the specific reality by acyclic Task Switching for periodic task in the present invention
Existing mode does not repeat them here.
S102, pending task-set is divided using space-division method, obtains first task subset.
Wherein, first task subset includes at least one pending task, and space-division method is used for by pending
The high-dimensional space of property parameters mapped of business is divided, to obtain the different arrangement groups of pending task spatially
It closes.
Optionally, property parameters include following one or more: when duty cycle, opposite deadline and the worst execution
Between.
In this step, by the high-dimensional space of property parameters mapped of pending task, i.e., for by pending task-set
Including the duty cycle of all pending tasks, the opposite deadline of task and task the worst execution time in one
Item or the high-dimensional space of multinomial mapped.Above-mentioned high-dimensional space is divided, multiple mutually independent scheduling sublayers are obtained
Space, and the corresponding scheduling sample of every sub-spaces, which is first task subset.
S103, the dispatch feature according to first task subset and preparatory trained schedulability decision model, acquisition can
Dispatch sex determination result, wherein schedulability determines result for indicating that can first task subset press in single core processor
It is scheduled according to EDF algorithm.
Specifically, dispatch feature is according in real-time system, and task-set is scheduled according to EDF algorithm in single core processor
When behavioural information and attribute information obtain, and the schedulability of task-set determines result tool in dispatch feature and real-time system
There is high correlation.Optionally, dispatch feature can be, but not limited to as one or more of: central processing unit (Central
Processing Unit, referred to as: CPU) utilization rate, density, superiority and inferiority degree, cutoff rate, implementation rate, scheduling rate, effectively scheduling
Rate is preempted rate, is effectively preempted rate, idleness, effective idleness, superperiod load factor, synchronizes busy cyclic loading rate, is special
Period load factor.It can describe in detail to the dispatch feature of task-set in subsequent embodiment.
In addition, the preparatory trained schedulability decision model employed in this step is a kind of based on data-driven
Schedulability Analysis model, the schedulability decision model are to be trained using deep learning method by great amount of samples, are learned
The real-time task collection dynamic dispatching process in EDF algorithm based on earliest Deadline First is practised, it is obtained.A kind of possibility
Implementation in, in the present embodiment, preparatory trained schedulability decision model is by a large amount of training samples, to depth
Neural network (Deep Neural Networks, referred to as: DNN) model is trained.
In actual application, the vector constituted using the dispatch feature of first task subset is schedulability test model
Input, the corresponding schedulability of first task subset determines result for the output of schedulability decision model.Using schedulable
Sex determination model carries out schedulability judgement, and obtained schedulability determines that the accuracy of result is higher.
In the present embodiment, firstly, pending task-set is obtained, then, using space-division method to pending task-set
It is divided, obtains first task subset, sentenced according to the dispatch feature of first task subset and preparatory trained schedulability
Cover half type obtains schedulability and determines as a result, schedulability determines result for indicating that can first task subset at monokaryon
It manages scheduled according to EDF algorithm on device.
The beneficial effect of method provided by the present embodiment is: can predict that can task-set using trained in advance
The schedulability decision model being scheduled in single core processor according to EDF algorithm carries out schedulability judgement, improves adjustable
Spend the accuracy of sex determination result.In addition, carrying out schedulability judgement to the corresponding first task subset of pending task-set, also
Decision-making foundation can be provided when complete pending task-set can not be scheduled.
On the basis of embodiment shown in Fig. 1, when complete pending task-set can not be scheduled, it may also include following
Step:
According to schedulability judgement as a result, can be scheduled in single core processor according to EDF algorithm, and pending task
The most first task subset of quantity is determined as goal task subset, so that single core processor is appointed according to EDF algorithmic dispatching target
Each pending task in subset of being engaged in.
Specifically, schedulability decision maker can determine result neatly according to the schedulability of all first task subsets
Select the first task subset that can be scheduled as goal task subset, so that single core processor is according to the EDF algorithmic dispatching mesh
Each pending task in mark task subset.
Preferably, by can in single core processor according to EDF algorithm be scheduled first task subset in, pending task
The most first task subset of quantity is determined as goal task subset, that is to say, that goal task subset is can be scheduled
One of them in one task subset.Wherein, if first task that can be scheduled in single core processor according to EDF algorithm
Collection, when the most first task subset of pending task quantity is unique, then the first task subset that this is uniquely met to condition is true
It is set to goal task subset；If can in single core processor according to EDF algorithm be scheduled first task subset in, pending
When the most first task subset of quantity of being engaged in is multiple, then wherein any one can be identified as goal task subset.
Schedulability decision maker can determine that result flexibly selects suitable first task subset to make according to schedulability
For goal task subset, the flexibility of scheduling is improved, and improves the scheduled probability of task.
Further, on the basis of embodiment shown in Fig. 1, step S102 be may be accomplished by: with pending
Task-set includes being illustrated for M pending tasks, wherein M is the integer greater than 0；
Fully intermeshing is carried out to the M pending tasks, is obtainedA first task subset.Its
In,Indicate the number that the I progress obtained first task subset of permutation and combination is chosen from M pending tasks, I is
Integer greater than 0, X are the number that M pending tasks are carried out with the first task subset that fully intermeshing obtains, and X is whole greater than 1
Number.
The mode of the pending task-set set is expressed as SM={ α1,...αb,...αM, wherein αbIt indicates b-th week
The pending task of phase property, preemptible, M are the integer greater than 0, and b is the integer greater than 0.
Firstly, by pending task-set SM={ α1,...αb,...αMBe mapped on metric space, formed one by respectively to
3 × M that the property parameters of execution task are constituted ties up metric space Z, wherein metric space Z can be indicated with formula (1):
Z=[[C1,D1,T1]T,...,[Cb,Db,Tb]T,...,[CM,DM,TM]T], Z ∈ R3×MFormula (1)
Wherein, C is the worst execution time of task, and D is the opposite deadline of task, and T is duty cycle, and should be wait hold
The value of the property parameters of each pending task is different from row task-set.
Then, fully intermeshing is carried out to M pending tasks in metric space, thus, by the high-dimensional metric space Z
It is divided intoIt is a by first task subsetThe mutually independent scheduling subspace constituted
ObtainA first task subset.
Wherein,Indicate first task subset,By in pending task-set SMIn appoint and take I pending tasks
Obtained according to priority descending sort, k indicates the number of the first task subset,Here, X indicates the number of first task subset.
Next, by a specific embodiment be discussed in detail employed in embodiment illustrated in fig. 1 it is preparatory it is trained can
How scheduling sex determination model obtains.
Fig. 2 is EDF schedulability determination method embodiment two flow diagram provided by the invention based on deep learning.
As shown in Fig. 2, method shown in the present embodiment includes:
S201, task-set is obtained, wherein task-set includes at least one task.
The task-set obtained in step s 201 includes all tasks that will appear in real-time system, these tasks all have
The periodical and property seized.
S202, task-set is divided using space-division method, obtains the second task subset.
Wherein, the second task subset carries schedulability judgement label, and schedulability label is for indicating the second task
Can subset scheduled according to EDF algorithm in single core processor, and the second task subset includes at least one task.
The purpose of this step is: building magnanimity training sample, so that DNN model can adequately be learnt.Therefore,
By carrying out space division to the task-set for including all tasks that will appear in real-time system, to obtain the second of substantial amounts
Task subset, the second task subset are the training sample of DNN model.Meanwhile also obtaining the adjustable of each second task subset
Spend sex determination label.
The schedulability of second task subset determines that label can determine according to the schedulability in EDF algorithm in the prior art
Formula obtains.
S203, label is determined according to the dispatch feature and schedulability of the second task subset, DNN model is instructed
Practice, obtains schedulability decision model.
The vector constructed by the dispatch feature of the second task subset and the corresponding schedulability of the second task subset are sentenced
Input of the calibration label as DNN model, so that DNN model carries out deep learning.By deep learning, continuous percentage regulation nerve
The precision of network model, to obtain to accurately identify and predicting that can task-set in single core processor according to EDF algorithm
Scheduled schedulability decision model.
In the present embodiment, task-set is divided using space-division method later firstly, obtaining task-set, is obtained
Second task subset, wherein the second task subset, which carries, indicates that can the second task subset in single core processor according to EDF
The scheduled schedulability of algorithm determines label, further, further according to the dispatch feature and schedulability of the second task subset
Determine label, DNN model is trained, obtains schedulability decision model.Method in the present embodiment passes through will be to derive
Mathematic(al) representation, building formalization schedulability decision condition be research conventional scheduling algorithms and with training sample, identify
Deep learning method for the purpose of mode combines, and obtains the higher schedulability decision model of accuracy.
Schedulability judgement is carried out to pending task-set using above-mentioned schedulability decision model, can not only improve can
Dispatch the accuracy of sex determination result.In addition it is possible to when complete pending task-set can not be scheduled, provide decision according to
According to improve the flexibility of real-time system scheduling and the probability that task is scheduled.
Fig. 3 is that the process of the EDF schedulability determination method embodiment three provided by the invention based on deep learning is illustrated
Figure.Embodiment illustrated in fig. 3 is for being described in detail how dispatch feature obtains.Method in embodiment illustrated in fig. 3 can be
It is executed before Fig. 1, embodiment illustrated in fig. 2.
As shown in figure 3, the method for the present embodiment includes:
S301, behavioural information and attribute information according to task when scheduled according to EDF algorithm in single core processor, structure
Build the first dispatch feature set, wherein the first dispatch feature set includes at least one first dispatch feature.
In this step, mainly through observation and analysis task collection when scheduled according to EDF algorithm in single core processor
Behavior and attribute, using constant feature and characteristics of variables, single order feature and multistage feature, gross feature and microscopic feature, dynamic
The means such as state feature and static nature, picked out from primitive scheduling data can portray principal contradiction, it is with physical significance
Data are dispatched, establish the first dispatch feature set according to the scheduling data picked out.
S302, according to preset condition, selected at least one first dispatch feature that the first dispatch feature set includes
Second dispatch feature.
The purpose of this step is that selecting from the first dispatch feature set can be such that schedulability test model obtains most
First dispatch feature of good performance, as the second dispatch feature.Second dispatch feature can be multiple, or one, the
Two dispatch features, which need to meet, determines that results relevance is strong with schedulability, the weak characteristic of correlation between the first dispatch feature.It is real
On border, the second dispatch feature is to meet the first dispatch feature of preset condition.
S303, data processing is carried out to the second dispatch feature, obtains dispatch feature.
Under normal conditions, there is the second dispatch feature higher-dimension characteristic to need to reduce calculation amount to the second dispatch feature
Carry out data processing.
Specifically, it by the processing such as the second dispatch feature being filtered and being denoised, second is adjusted to obtain that treated
Spend feature, wherein there is the second dispatch feature that treated higher schedulability to determine accuracy, and have low-dimensional characteristic.It will
The second dispatch feature that treated, as the dispatch feature for characterizing the judgement of task-set schedulability.
Illustratively, dispatch feature may include: cpu busy percentage, density, superiority and inferiority degree, cutoff rate, implementation rate, scheduling
Rate, effectively scheduling rate are preempted rate, are effectively preempted rate, idleness, effective idleness, superperiod load factor, synchronize the busy period
Load factor, special period load factor etc., it should be noted that above-mentioned dispatch feature indicates the dispatch feature of task-set.
The dispatch feature of task-set is explained in detail below:
Assuming that task-set SnIncluding n task, wherein τiFor task-set SnIn i-th of task, wherein n be greater than 0
Integer, i is the integer less than or equal to n, in described in detail below, CiIndicate task τiThe worst execution time, DiIt indicates to appoint
Be engaged in τiOpposite deadline, TiIndicate task τiDuty cycle.
SnCPU resource utilization: indicate SnIn each task the sum of CPU resource utilization, be denoted as U (Sn),The CPU resource utilization of task-set is bigger, is less susceptible to be scheduled.
Wherein, UiIndicate τiCPU resource utilization, refer to τiThe worst execution time and period ratio,
UiTo describe the case where task is using CPU resource utilization, UiBigger, the resource utilization of task CPU is higher, which gets over
Hardly possible is scheduled.
SnDensity: indicate SnIn each task the sum of density, be denoted as δ (Sn),
Task denseness of set is bigger, is less susceptible to be scheduled.
Wherein, δiIndicate τiTask density, refer to τiThe worst execution time and the worst deadline ratio,To describe density degree of the execution time of task within opposite deadline, δiIt is bigger, illustrate that the task is got over
Hardly possible is scheduled.
SnSuperiority and inferiority degree: indicate SnIn high-quality task number and total task number ratio, be denoted as Φ (Sn),Wherein NΦIIIndicate high-quality task number, n indicates SnThe sum of middle task.As Φ (Sn)≥ΦIIWhen, SnQuilt
It is known as high-quality task-set, otherwise, task-set inferior of being known as.Here ΦIIIt is initialized to 50%.When high-quality in task-set
Task number is more, and the task-set is easier to be scheduled.That is, the superiority and inferiority degree of task-set is the scheduled hardly possible of task-set
Easy degree.
Wherein, whether task is high-quality task, can be determined in the following manner:
ΦiIndicate τiSuperiority and inferiority degree,Work as Φi> ΦIWhen, τiReferred to as excellent task, works as Φi< ΦI
When, τiReferred to as bad task.ΦIPreset threshold value is indicated, by as the index for measuring Task Quality.So-called high-quality task
Refer in one cycle, the worst execution time of task is relatively short but deadline and the period relatively is all longer, makes task
There is time enough to execute, more good task is easier to be scheduled, the build-in attribute of the task of the superiority and inferiority degree of task.
SnCutoff rate: indicate SnIn maximum absolute cutoff time and superperiod of each task within its superperiod ratio,
It is denoted as Q (Sn), whereindi(Tlcm(Sn)) indicate task τiIn superperiod Tlcm
(Sn) in last time scheduling the absolute cutoff time, di(Tlcm(Sn))=Tlcm(Sn)-Ti+Di.When the cutoff rate of task-set is got over
Greatly, task-set is limited smaller, easier to be scheduled, the superperiod T of task-setlcm(Sn) indicate a duty cycle in task-set
Least common multiple.
SnImplementation rate: indicate SnIn each task executions time total within its superperiod and the ratio for finally completing the time,
It is denoted as J (Sn), whereinHere, Ri(Tlcm(Sn)) indicate task τiIn superperiod Tlcm
(Sn) in last time scheduling worst-case response time, Ri(Tlcm(Sn))=Tlcm(Sn)-Ti+Ri(Sn).When the execution of task-set
Rate is bigger, and the degree that task-set executes is higher, easier to be scheduled.
SnScheduling rate: indicate SnIn the ratio that finally completes time and superperiod of each task within its superperiod, be denoted as
F(Sn), whereinWhen the possibility that the scheduling rate of task-set is bigger, and task-set is scheduled
Property is smaller, is more difficult to be scheduled.
SnEffective scheduling rate: indicate SnIn each task finally complete time and maximum deadline within its superperiod
Ratio, be denoted as EF (Sn), whereinWhen effective scheduling rate of task-set
Bigger, a possibility that task-set is overtime when scheduled, is bigger, is more difficult to be scheduled.
SnBe preempted rate: indicate SnIn each task the maximum value for being preempted rate, be denoted as Z (Sn), whereinIt is preempted that rate is bigger, and task influences each other bigger in task-set when task-set, is more difficult to be adjusted
Degree.
Wherein, ZiIndicate τiIn SnIn be preempted rate, refer to SnIn τiThe maximum time being preempted at the end of execution
With the ratio in period, wherein the maximum time that task is preempted at the end of execution is equal to the worst-case response time R of taskiWith
The worst execution time C of taskiDifference, then,Task is preempted rate ZiTo describe task in the period
The degree being inside preempted, ZiIt is bigger, task by within the period other task influence degrees it is higher, be more difficult to be scheduled.
SnBe effectively preempted rate: indicate SnIn each task be effectively preempted the maximum value of rate, be denoted as EZ (Sn), whereinEffectively it is preempted that rate is bigger, and task influences each other more in task-set when task-set
Greatly, it is more difficult to be scheduled.
Wherein, EZi(Sn) indicate τiIn SnIn effectively be preempted rate, refer to SnIn τiIt is preempted at the end of execution
The ratio of maximum time and opposite deadline, wherein the maximum time that task is preempted at the end of execution is equal to task
Worst-case response time RiWith the worst execution time C of taskiDifference, then,The effective of task is robbed
Account for rate EZi(Sn) degree that is preempted within opposite deadline to describe task, EZi(Sn) bigger, task is by within the period
Other task influence degrees are higher, are more difficult to be scheduled.
SnIdleness: indicate SnIn each task idleness minimum value, be denoted as B (Sn), whereinWhen the idleness of task-set is bigger, task idle degrees are higher in task-set, easier to be adjusted
Degree.
Wherein, Bi(Sn) indicate τiIn SnIn idleness, refer to SnIn τiAt it with respect to remaining time in deadline
With the ratio in period, wherein SnIn τiIt is equal to the opposite deadline D of task with respect to remaining time in deadline at itiWith
The worst-case response time R of taskiDifference, then,The idleness B of taski(Sn) to describe task
The idle degrees within the period, Bi(Sn) it is bigger, task by within the period idle degrees it is higher, it is easier to be scheduled.
SnEffective idleness: indicate SnIn each task idleness minimum value, be denoted as EB (Sn), whereinWhen effective idleness of task-set is bigger, free time of the task-set within opposite deadline
Degree is higher, easier to be scheduled.
Wherein, EBi(Sn) indicate τiIn SnIn effective idleness, refer to SnIn τiAt it with respect to surplus in deadline
The ratio of remaining time and opposite deadline, wherein SnIn τiIt is equal to the phase of task with respect to remaining time in deadline at it
To deadline DiWith the worst-case response time R of taskiDifference, then,To describe task in week
Idle degrees in phase,It is bigger, task by within opposite deadline idle degrees it is higher, it is easier to be scheduled.
SnSuperperiod load factor: indicate SnCpu load within its superperiod and the ratio in synchronous busy period, are denoted as LF
(Tlcm(Sn)),Wherein, Tlcm(Sn) indicate task-set SnSuperperiod.
SnSynchronize busy cyclic loading rate: indicate SnCpu load in the busy period and the ratio in synchronous busy period are synchronized at it
Value, is denoted as LF (BL (Sn)),BL(Sn) indicate SnSynchronize the busy period.
SnSpecial period load factor: indicate SnIn special period Lmax(Sn) in cpu load and the special time
The ratio of section, is denoted as LF (Lmax(Sn)),Lmax(Sn) indicate judging task-set Sn
A time range upper bound is needed to be traversed for when can be by EDF algorithmic dispatching.
In the present embodiment, behavioural information first according to task when scheduled according to EDF algorithm in single core processor and
Attribute information constructs the first dispatch feature set, wherein and the first dispatch feature set includes at least one first dispatch feature,
Then, according to preset condition, the second scheduling is selected at least one first dispatch feature that the first dispatch feature set includes
Feature further carries out data processing to the second dispatch feature, obtains dispatch feature.The institute of method shown in through this embodiment
The dispatch feature of acquisition determines that result has high correlation with schedulability, can be improved the standard of schedulability decision model
Exactness.
Fig. 4 is the structural representation of the EDF schedulability decision maker embodiment one provided by the invention based on deep learning
Figure.As shown in figure 4, the device 40 includes: to obtain module 41, division module 42 and determination module 43.
Wherein, module 41 is obtained, for obtaining pending task-set, wherein pending task-set includes that at least one is waited for
Execution task.
Division module 42 obtains first task for dividing using space-division method to pending task-set
Collection, wherein first task subset includes at least one pending task, and space-division method is indicated to the category by pending task
The property high-dimensional space of parameter mapped is divided, and the different permutation and combination of pending task spatially are obtained.
Determination module 43, for determining mould according to the dispatch feature and preparatory trained schedulability of first task subset
Type obtains schedulability and determines result, wherein schedulability determines that result indicates that can first task subset in single core processor
On it is scheduled according to EDF algorithm.
Optionally, above-mentioned property parameters include following one or more: duty cycle, opposite deadline and the worst being held
The row time.
Optionally, dispatch feature may include cpu busy percentage, density, superiority and inferiority degree, cutoff rate, implementation rate, scheduling rate,
Effective scheduling rate, be preempted rate, be effectively preempted rate, idleness, effective idleness, superperiod load factor, to synchronize the busy period negative
Load rate, special period load factor etc..
Device shown in the present embodiment can be used for executing method shown in FIG. 1, and it is similar that the realization principle and technical effect are similar, this
Place repeats no more.
Optionally, pending task includes M pending tasks, and M is the integer greater than 0；
Correspondingly, division module 42 are specifically used for carrying out fully intermeshing to M pending tasks, obtainA first task subset, whereinIndicate that choosing I from M pending tasks carries out arrangement group
The number of obtained first task subset is closed, I is the integer greater than 0, and X is to carry out fully intermeshing to M pending tasks to obtain
First task subset number, X is integer greater than 1.
In some embodiments, which can also comprise determining that module 44.
Wherein it is determined that module 44, for being determined according to schedulability as a result, can be calculated in single core processor according to EDF
Method is scheduled, and the first task subset of pending task quantity at most is determined as goal task subset, so that at monokaryon
Device is managed according to each pending task in EDF algorithmic dispatching goal task subset.
Fig. 5 is the structural representation of the EDF schedulability decision maker embodiment two provided by the invention based on deep learning
Figure.As shown in figure 5, the device 50 of the present embodiment is on the basis of the embodiment shown in fig. 4, further includes: training module 45.
Correspondingly, module 41 is obtained, is also used to obtain task-set, wherein task-set includes at least one task.
Division module 42 is also used to divide task-set using space-division method, obtains the second task subset,
In, the second task subset carries schedulability and determines that label, schedulability determine label for indicating the second task subset energy
No scheduled according to EDF algorithm in single core processor, the second task subset includes at least one task.
Training module 45, for determining label according to the dispatch feature and schedulability of the second task subset, to DNN mould
Type is trained, and obtains schedulability decision model.
Device shown in the present embodiment can be used for executing the technical solution in embodiment illustrated in fig. 2, realization principle and skill
Art effect is similar, and details are not described herein again.
Fig. 6 is the structural representation of the EDF schedulability decision maker embodiment three provided by the invention based on deep learning
Figure.As shown in fig. 6, on the basis of the embodiment shown in Fig. 5 of device 60 of the present embodiment, further includes: extraction module 46.
Extraction module 46, for extracting dispatch feature.
In some embodiments, extraction module 46 includes: building submodule 461, selection submodule 462 and processing submodule
Block 463.
Submodule 461 is constructed, specifically for the behavior according to task when scheduled according to EDF algorithm in single core processor
Information and attribute information construct the first dispatch feature set, wherein the first dispatch feature set includes at least one first scheduling
Feature.
Submodule 462 is selected, is specifically used for according to preset condition, at least one that the first dispatch feature set includes the
The second dispatch feature is selected in one dispatch feature.
Submodule 463 is handled, is specifically used for carrying out data processing to the second dispatch feature, obtains dispatch feature.
Device shown in the present embodiment can be used for executing the technical solution in embodiment illustrated in fig. 3, realization principle and skill
Art effect is similar, and details are not described herein again.
Fig. 7 is the structural representation of the EDF schedulability decision maker example IV provided by the invention based on deep learning
Figure.As shown in fig. 7, device 70 shown in the present embodiment includes: memory 71, processor 72.
Memory 71 can be independent physical unit, can be connect by bus 53 with processor 72.Memory 71, place
Reason device 72 also can integrate together, pass through hardware realization etc..
Memory 71 realizes above method embodiment for storing, and processor 72 calls the program, and it is real to execute above method
Apply the operation of example.
Optionally, when passing through software realization some or all of in the method for above-described embodiment, above-mentioned apparatus 70 can also
To only include processor 72.Memory 71 for storing program is located at except device 70, processor 72 by circuit/electric wire with
Memory connection, for reading and executing the program stored in memory.
Processor 72 can be central processing unit (Central Processing Unit, CPU), network processing unit
The combination of (Network Processor, NP) or CPU and NP.
Processor 72 can further include hardware chip.Above-mentioned hardware chip can be specific integrated circuit
(Application-Specific Integrated Circuit, ASIC), programmable logic device (Programmable
Logic Device, PLD) or combinations thereof.Above-mentioned PLD can be Complex Programmable Logic Devices (Complex
Programmable Logic Device, CPLD), field programmable gate array (Field-Programmable Gate
Array, FPGA), Universal Array Logic (Generic Array Logic, GAL) or any combination thereof.
Memory 71 may include volatile memory (Volatile Memory), such as random access memory
(Random-Access Memory, RAM)；Memory also may include nonvolatile memory (Non-volatile
), such as flash memory (Flash Memory), hard disk (Hard Disk Drive, HDD) or solid state hard disk Memory
(Solid-state Drive, SSD)；Memory can also include the combination of the memory of mentioned kind.
The present invention also provides a kind of program products, for example, computer storage medium, comprising: program, program is by processor
For executing above method when execution.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above-mentioned each method embodiment can lead to
The relevant hardware of program instruction is crossed to complete.Program above-mentioned can be stored in a computer readable storage medium.The journey
When being executed, execution includes the steps that above-mentioned each method embodiment to sequence；And storage medium above-mentioned include: ROM, RAM, magnetic disk or
The various media that can store program code such as person's CD.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (10)
1. a kind of EDF schedulability determination method based on deep learning characterized by comprising
Obtain pending task-set, wherein the pending task-set includes at least one periodically and preemptible is pending
Task；
The pending task-set is divided using space-division method, obtains first task subset, wherein described first
Task subset includes at least one described pending task, and the space-division method is used for the category by the pending task
The property high-dimensional space of parameter mapped is divided, to obtain the different permutation and combination of the pending task spatially；
According to the dispatch feature of the first task subset and preparatory trained schedulability decision model, schedulability is obtained
Determine result, wherein the schedulability determines result for indicating that can the first task subset in single core processor
It is scheduled according to earliest Deadline First EDF algorithm.
2. the method according to claim 1, wherein the dispatch feature according to the first task subset and
Preparatory trained schedulability decision model, acquisition schedulability determine after result, further includes:
According to schedulability judgement as a result, can be scheduled in single core processor according to EDF algorithm, and pending task
The most first task subset of quantity is determined as goal task subset, so that single core processor is according to EDF algorithmic dispatching institute
State each pending task in goal task subset.
3. the method according to claim 1, wherein the pending task-set includes M described pending
Business, M are the integer greater than 0；
It is described that the pending task-set is divided using space-division method, obtain first task subset, comprising:
Fully intermeshing is carried out to the M pending tasks, is obtainedA first task subset,
In,Indicate the number that the I progress obtained first task subset of permutation and combination is chosen from the M pending tasks,
I is the integer greater than 0, and X is the number that the M pending tasks are carried out with the first task subset that fully intermeshing obtains, and X is big
In 1 integer.
4. the method according to claim 1, wherein the trained schedulability decision model in advance passes through
Following manner obtains:
Obtain task-set, wherein the task-set includes at least one task；
The task-set is divided using space-division method, obtains the second task subset, wherein the second task
Collection carries schedulability and determines that label, the schedulability determine label for indicating that can the second task subset in list
Scheduled according to EDF algorithm on core processor, the second task subset includes at least one described task；
Label is determined according to the dispatch feature of the second task subset and the schedulability, to deep neural network DNN mould
Type is trained, and obtains the schedulability decision model.
5. the method according to claim 1, wherein the dispatch feature is extracted in the following manner:
According to behavioural information and attribute information of the task when scheduled according to EDF algorithm in single core processor, building first is adjusted
Spend characteristic set, wherein the first dispatch feature set includes at least one first dispatch feature；
According to preset condition, is selected at least one described first dispatch feature that the first dispatch feature set includes
Two dispatch features；
Data processing is carried out to second dispatch feature, obtains the dispatch feature.
6. method according to claim 1-5, which is characterized in that the property parameters include with the next item down or more
:
Duty cycle, opposite deadline and the worst execution time.
7. method according to claim 1-5, which is characterized in that the dispatch feature includes with the next item down or more
:
Central processor CPU utilization rate, density, superiority and inferiority degree, cutoff rate, implementation rate, scheduling rate, effectively scheduling rate, be preempted
Rate, be effectively preempted rate, idleness, effective idleness, superperiod load factor, to synchronize busy cyclic loading rate, special period negative
Load rate.
8. a kind of EDF schedulability decision maker based on deep learning characterized by comprising
Module is obtained, for obtaining pending task-set, wherein the pending task-set includes at least one pending
Business；
Division module, for being divided using space-division method to the pending task-set, acquisition first task subset,
Wherein, the first task subset includes at least one described pending task, and the space-division method is indicated to by described
The high-dimensional space of property parameters mapped of pending task is divided, and obtains the pending task spatially not
Same permutation and combination；
Determination module, for determining mould according to the dispatch feature and preparatory trained schedulability of the first task subset
Type obtains schedulability and determines result, wherein the schedulability determines that result indicates that can the first task subset in list
It is scheduled according to earliest Deadline First EDF algorithm on core processor.
9. a kind of EDF schedulability decision maker based on deep learning characterized by comprising memory and processor；
The memory stores program instruction；
The processor executes described program instruction, requires method described in any one of 1-7 with perform claim.
10. a kind of storage medium characterized by comprising program, described program is when being executed by processor, with perform claim
It is required that method described in any one of 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811486770.6A CN109739638A (en) | 2018-12-06 | 2018-12-06 | A kind of EDF schedulability determination method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811486770.6A CN109739638A (en) | 2018-12-06 | 2018-12-06 | A kind of EDF schedulability determination method and device based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109739638A true CN109739638A (en) | 2019-05-10 |
Family
ID=66358510
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811486770.6A Pending CN109739638A (en) | 2018-12-06 | 2018-12-06 | A kind of EDF schedulability determination method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109739638A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112416566A (en) * | 2020-05-19 | 2021-02-26 | 中国航空无线电电子研究所 | IMA general processing module resource scheduling analysis method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103473120A (en) * | 2012-12-25 | 2013-12-25 | 北京航空航天大学 | Acceleration-factor-based multi-core real-time system task partitioning method |
| CN104834556A (en) * | 2015-04-26 | 2015-08-12 | 西北工业大学 | Mapping method for multimode real-time tasks and multimode computing resources |
| CN105700948A (en) * | 2014-11-24 | 2016-06-22 | 阿里巴巴集团控股有限公司 | Method and device for scheduling calculation task in cluster |
| CN106125901A (en) * | 2016-08-16 | 2016-11-16 | 华中科技大学 | A kind of dynamic energy-saving real-time scheduling method based on interval division |
| CN107193649A (en) * | 2017-05-25 | 2017-09-22 | 深信服科技股份有限公司 | A kind of method for scheduling task and device based on NUMA system |
| US20170286180A1 (en) * | 2016-03-31 | 2017-10-05 | International Business Machines Corporation | Joint Network and Task Scheduling |
| CN108335012A (en) * | 2017-12-26 | 2018-07-27 | 佛山科学技术学院 | A kind of intelligence remote sensing satellite stratification distributed freedom cotasking planning system |
| CN108717572A (en) * | 2018-05-31 | 2018-10-30 | 佛山科学技术学院 | Imaging task schedulability prediction technique based on structural neural networks |
-
2018
- 2018-12-06 CN CN201811486770.6A patent/CN109739638A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103473120A (en) * | 2012-12-25 | 2013-12-25 | 北京航空航天大学 | Acceleration-factor-based multi-core real-time system task partitioning method |
| CN105700948A (en) * | 2014-11-24 | 2016-06-22 | 阿里巴巴集团控股有限公司 | Method and device for scheduling calculation task in cluster |
| CN104834556A (en) * | 2015-04-26 | 2015-08-12 | 西北工业大学 | Mapping method for multimode real-time tasks and multimode computing resources |
| US20170286180A1 (en) * | 2016-03-31 | 2017-10-05 | International Business Machines Corporation | Joint Network and Task Scheduling |
| CN106125901A (en) * | 2016-08-16 | 2016-11-16 | 华中科技大学 | A kind of dynamic energy-saving real-time scheduling method based on interval division |
| CN107193649A (en) * | 2017-05-25 | 2017-09-22 | 深信服科技股份有限公司 | A kind of method for scheduling task and device based on NUMA system |
| CN108335012A (en) * | 2017-12-26 | 2018-07-27 | 佛山科学技术学院 | A kind of intelligence remote sensing satellite stratification distributed freedom cotasking planning system |
| CN108717572A (en) * | 2018-05-31 | 2018-10-30 | 佛山科学技术学院 | Imaging task schedulability prediction technique based on structural neural networks |
Non-Patent Citations (1)
| Title |
|---|
| 张永悦: "基于分区的航电系统调度分析工具实现", 《计算机工程》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112416566A (en) * | 2020-05-19 | 2021-02-26 | 中国航空无线电电子研究所 | IMA general processing module resource scheduling analysis method |
| CN112416566B (en) * | 2020-05-19 | 2024-02-27 | 中国航空无线电电子研究所 | Analysis method for IMA universal processing module resource scheduling |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Vaz et al. | PSwarm: a hybrid solver for linearly constrained global derivative-free optimization | |
| JP6384065B2 (en) | Information processing apparatus, learning method, and program | |
| US11915101B2 (en) | Numerical quantum experimentation | |
| CN108122032A (en) | A kind of neural network model training method, device, chip and system | |
| CN103593323A (en) | Machine learning method for Map Reduce task resource allocation parameters | |
| CN113220450B (en) | Load prediction method, resource scheduling method and device for cloud-side multi-data center | |
| Chen et al. | Optimal variability sensitive condition-based maintenance with a Cox PH model | |
| CN105868222A (en) | Task scheduling method and device | |
| CN111950810B (en) | Multi-variable time sequence prediction method and equipment based on self-evolution pre-training | |
| CN114895773A (en) | Energy consumption optimization method, system and device of heterogeneous multi-core processor and storage medium | |
| CN112052081A (en) | Task scheduling method and device and electronic equipment | |
| CN111198754A (en) | Task scheduling method and device | |
| CN109766181A (en) | A kind of RMS schedulability determination method and device based on deep learning | |
| CN109739638A (en) | A kind of EDF schedulability determination method and device based on deep learning | |
| CN109063128A (en) | Integrated Density Estimator device window parameter optimization method, device and terminal device | |
| CN112598443A (en) | Online channel business data processing method and system based on deep learning | |
| CN116306030A (en) | New energy prediction dynamic scene generation method considering prediction error and fluctuation distribution | |
| Li et al. | The extreme counts: modeling the performance uncertainty of cloud resources with extreme value theory | |
| CN110443282A (en) | A kind of Embryonic Stages classification method in embryo's timing image | |
| CN114021733B (en) | Model training optimization method, device, computer equipment and storage medium | |
| Hu et al. | A greedy-based crow search algorithm for semiconductor final testing scheduling problem | |
| WO2022142026A1 (en) | Classification network construction method, and classification method based on classification network | |
| CN108471362B (en) | Resource allocation prediction technique and device | |
| Baughman et al. | Measuring, quantifying, and predicting the cost-accuracy tradeoff | |
| Jananee et al. | Allocation of cloud resources based on prediction and performing auto-scaling of workload |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190510 |