CN116541154B - Intelligent medical-oriented personalized application scheduling method and device - Google Patents
Intelligent medical-oriented personalized application scheduling method and device Download PDFInfo
- Publication number
- CN116541154B CN116541154B CN202310825753.5A CN202310825753A CN116541154B CN 116541154 B CN116541154 B CN 116541154B CN 202310825753 A CN202310825753 A CN 202310825753A CN 116541154 B CN116541154 B CN 116541154B
- Authority
- CN
- China
- Prior art keywords
- power consumption
- qos
- application program
- application
- delay
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 230000004044 response Effects 0.000 claims abstract description 30
- 238000012545 processing Methods 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000010801 machine learning Methods 0.000 claims description 8
- 238000005265 energy consumption Methods 0.000 claims description 5
- 230000003111 delayed effect Effects 0.000 claims description 4
- 238000012986 modification Methods 0.000 claims description 4
- 230000004048 modification Effects 0.000 claims description 4
- 230000002068 genetic effect Effects 0.000 claims description 3
- 230000001172 regenerating effect Effects 0.000 claims description 2
- 238000013461 design Methods 0.000 abstract description 6
- 238000012549 training Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000006872 improvement Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000005192 partition Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000000638 solvent extraction Methods 0.000 description 3
- 238000005382 thermal cycling Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 241000257303 Hymenoptera Species 0.000 description 1
- 241001206881 Myrmeleon inconspicuus Species 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000008646 thermal stress Effects 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/40—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management of medical equipment or devices, e.g. scheduling maintenance or upgrades
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Business, Economics & Management (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a personalized application scheduling method and device for intelligent medical treatment, comprising the following steps: step S1, acquiring the load of each application program; step S2, obtaining preference of a single intelligent medical application program for response delay and power consumption and QoS under delay power consumption level according to the load of each application program; and step S3, maximizing the QoS of the personalized intelligent medical application program according to the preference of the single intelligent medical application program for response delay and power consumption and the QoS under the delay power consumption level. By adopting the technical scheme of the invention, the parallel computing capability of the coprocessor can be fully exerted under the condition of meeting all design constraint conditions such as time, temperature, service life reliability, energy budget and the like, and the QoS maximization of the application program is realized.
Description
Technical Field
The invention belongs to the technical field of intelligent medical treatment, and particularly relates to a personalized application scheduling method and device for intelligent medical treatment.
Background
In recent years, rapid advances in information technology have prompted the widespread deployment of medical information physical systems (MCPS), particularly in the field of digital medicine. In the field of intelligent medical science, a CPU-GPU (graphics processing unit) cooperates with a multiprocessor system on a chip (MPSoC) enabled medical edge device to fully exploit the parallel computing capability of the GPU core and the general computing capability of the CPU core, which has great advantages in terms of processing and managing massive health-related data, but faces the key challenge of mapping and partitioning application workloads onto the CPU and GPU cores, i.e. how application workload partitions determine the appropriate proportion of application workloads between the CPU and GPU cores. However, existing research often uses worst-case execution cycles to pessimistically predict the workload of an individual application, resulting in a significant reduction in application quality of service (QoS). The experience prediction technology is developed rapidly, and becomes a feasible technology for predicting the workload of medical application programs, but the problem of scarcity of training sample data exists in the early system deployment stage, and higher estimation accuracy cannot be maintained. Therefore, establishing an efficient workload prediction method with high estimation accuracy but requiring only a small number of training samples has become a very important research topic.
In the past, intelligent medical research in the MCPS environment has become a hot topic, a great deal of scholars' research work is focused on processing the application mapping and partitioning problems of the CPU-GPU co-processor, namely, optimizing the response delay and energy consumption of the co-processor, but the work is only to carry out the algorithm design of delay and energy consumption from the system point of view, namely, the algorithms are formulated for improving the average performance of the whole application, and neglecting the personalized requirements of the digital medical application in the MCPS field. Furthermore, CPU and GPU core life reliability indicators are important indicators to quantify semiconductor devices' endurance to permanent errors caused by wear-out failures, but existing work does not take CPU-GPU core life reliability into account, resulting in hardware failures occurring ahead of time. Moreover, due to heterogeneity in lifetime reliability of the CPU core and the GPU core, unbalanced corruption between different cores may eventually lead to early failure of the system.
Disclosure of Invention
The invention aims to solve the technical problem of providing a personalized application scheduling method and device for intelligent medical treatment, which fully exert the parallel computing capacity of a co-processor and realize the QoS maximization of an application program under the condition of meeting all design constraint conditions such as time, temperature, service life reliability, energy budget and the like.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
A personalized application scheduling method for intelligent medical treatment comprises the following steps:
step S1, acquiring the load of each application program;
Step S2, obtaining preference of a single intelligent medical application program for response delay and power consumption and QoS under delay power consumption level according to the load of each application program;
And step S3, maximizing the QoS of the personalized intelligent medical application program according to the preference of the single intelligent medical application program for response delay and power consumption and the QoS under the delay power consumption level.
Preferably, in step S1, the load of each application is predicted based on machine learning.
Preferably, in step S2, individual smart medical applications are predicted for response delay and power consumption preferences and QoS at multiple delay power consumption levels based on the personality-driven applications;
preferably, in step S3, qoS of the personalized smart medical application is maximized based on the c-DPEA group smart application scheduling policy.
The invention also provides a personalized application scheduling device oriented to intelligent medical treatment, which comprises the following steps:
the acquisition module is used for acquiring the load of each application program;
The first processing module is used for obtaining the preference of the single intelligent medical application program for response delay and power consumption and QoS under the delay power consumption level according to the load of each application program;
And the second processing module is used for maximizing the QoS of the personalized intelligent medical application program according to the preference of the single intelligent medical application program for response delay and power consumption and the QoS under the delay power consumption level.
Preferably, the acquisition module predicts the load of each application program through machine learning.
Preferably, the first processing module predicts preferences of individual smart medical applications for response delay and power consumption and QoS at multiple delay power consumption levels based on personality driven applications;
Preferably, the second processing module maximizes QoS of the personalized smart medical application based on the c-DPEA community smart application scheduling policy.
Compared with the prior art, the invention has the following advantages and technical effects:
The invention provides a personalized scheduling technology for guaranteeing reliability for intelligent medical application program energized by CPU-GPU collaborative MPSoC through an application program load prediction method based on machine learning, an application program QoS prediction method based on personalized driving and an intelligent application program scheduling strategy based on c-DPEA group. The invention can maximize QoS of the personalized intelligent medical application program on the premise of meeting all design constraints.
Drawings
In order to more clearly illustrate the technical solutions of the present invention, the drawings that are needed in the embodiments are briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a method according to an embodiment of the invention;
FIG. 2 is a schematic diagram of a confusion matrix for the training process of the QoS prediction method of the personality driver according to the present invention;
FIG. 3 is a generalized capability diagram of the personality-driven QoS prediction method of the present invention for applications FDEB, FIR, KNN and EP;
FIG. 4 is a schematic diagram of the comparison of the c-DPEA-based application scheduling strategy developed in the present invention with five exemplary non-dominant solution search methods NSGA-III, MOEA/D, DLS-MOEA, maOEA/IGD, and MoALO;
FIG. 5 is a diagram of QoS comparison results obtained by the solution of the present invention and benchmark solution TEEM when running different applications;
Fig. 6 is a schematic diagram of a comparison of scheduling feasibility of the solution of the present invention and baseline test scheme TEEM.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
As shown in fig. 1, the personalized application scheduling method for intelligent medical treatment provided by the invention comprises the following steps:
A personalized application scheduling method for intelligent medical treatment comprises the following steps:
step S1, predicting the load of each application program based on machine learning;
Step S2, predicting the preference of a single intelligent medical application program for response delay and power consumption and QoS under various delay power consumption levels based on the individual driving application program according to the load of each application program;
And step S3, according to the preference of the single intelligent medical application program for response delay and power consumption and QoS under the delay power consumption level, maximizing the QoS of the personalized intelligent medical application program based on the c-DPEA group intelligent application program scheduling strategy.
In order to overcome the disadvantages of the conventional empirical prediction technology, the step S1 has the capability of maintaining high estimation accuracy but only requiring a small number of training samples, the step S2 considers that the preference of the application program for response delay and energy consumption is highly related to the program characteristics thereof, and the step S3 searches for the non-dominant scheduling solution with maximized QoS in order to find the optimal application program mapping and partitioning point with maximized QoS.
As an embodiment of the implementation of the present invention, step S1 specifically includes:
Step A1: random initialization parameters ；
Step A2: sample set of application programsDivided into meta training setsSum meta-test set/>Wherein/>Is the total number of application programs;
Step A3: at the position of Upper initialization mean square error/>The method comprises the following steps: /(I)；
Step A4: will mark the bitThe value is 1, namely: /(I)；
Step A5: from the slaveIn a total of/>Sample subset/>Wherein/>And/>From/>, respectivelyObtained/>Input values and output values;
Step A6: using the formula At/>Upper calculation; Wherein/>For the band parameter vector/>Regression model of/>，/>Is a partial derivative function;
step A7: updating parameters ，/>Wherein/>Is parameterized vector/>，/>Is a constant learning rate;
Step A8: constructing a new sample subset Performing meta-updating; wherein/>Belonging to/>But not belong to/>；
Step A9: updatingThe formula is: /(I)；
Step A10: judgingWhether or not it is. If yes, turning to the step A5; otherwise, turning to step A11;
step A11: in the book subset On performing meta-update correction/>The formula is: ; wherein/> Taking the step length as a primitive step length;
step A12: recording optimal parameter vector The formula is: /(I)；
Step A13: updatingOn/>；
Step A14: judgingWhether or not the specified threshold mean square error/>I.e. judgingIf so, turning to the step 4; otherwise, turning to step A15; wherein/>Is a constant with small numerical value and is used for judging/>And/>Whether or not it is sufficiently close;
Step A15: after the load of the application program is estimated, the best parameter vector is returned Regression model of/>And (5) exiting.
As an embodiment of the implementation of the present invention, step S2 specifically includes:
Step B1: initializing loop flags ; Wherein/>The number of applications to participate in the questionnaire;
Step B2: initializing preference weights ；
Step B3: using the formulaComputing application/>In/>Response delay/>, at a single delay power consumption levelWherein/>Is/>Start time of/>Is/>At execution time on CPU core only,/>Is/>Execution time on GPU kernel only,/>Is/>Proportion of workload on CPU and GPU cores,/>Is the total delay energy level;
Step B4: using the formula Calculation/>In/>Power consumption at a delayed power consumption level/>Wherein/>Is the power consumption of CPU,/>Is the power consumption of the GPU;
Step B5: using the formula Calculating latent variables/>Finger/>Preference for response delay and power consumption;
Step B6: using the formula Calculation ofIn/>QoS values at the individual delay energy levels/>Wherein/>And/>Is a decision variable;
step B7: solving the formula Calculating decision variables/>And/>Optimum value/>And/>；
Step B8: will mark the bitThe value is 1, namely: /(I)；
Step B9: solving the formulaCalculation/>Optimum value of (2)；
Step B10: comparing randomly generated preference weightsAnd calculated optimum/>Absolute difference/>；
Step B11: determining absolute differenceIf the value is smaller than the preset value, turning to the step B12; otherwise, turning to step B15;
Step B12: updating The formula is: /(I)；
Step B13: updatingThe formula is: /(I)；
Step B14: judgingIf so, turning to the step B9; otherwise, turning to step B15;
Step B15: regenerating preference weights ；
Step B16: judgingIf so, turning to the step B3; otherwise, turning to step B17;
Step B17: solving the formula Calculating a parameter matrix/>Optimum value/>; Wherein,For preference weight matrix, the formula is: /(I)； />For dimension/>Is a feature score matrix of (a);
Step B18: will mark the bit The value is 1, namely: /(I)；
Step B19: acquisition ofFeature score/>；
Step B20: using the formulaCalculation/>At/>Delaying QoS at the power consumption level;
Step B21: by the formula Calculating the optimal value/>；
Step B22: updatingThe formula is: /(I)；
Step B23: judgingIf so, turning to the step B19; otherwise, go to step B24; wherein/>Is the total number of applications.
Step B24: after application program preference and QoS are calculated, preference weight is returnedDecision variablesAnd/>QoS for applications at delayed energy consumption levels, i.e./>。
As an embodiment of the implementation of the present invention, step S3 specifically includes:
Step C1: predicting an application set using the application load prediction method described in step S1 A workload of each application;
step C2: initializing an iteration counter I.e./>；
Step C3: random initialization of two populationsAnd/>；
Step C4: judgingIf so, turning to step C5 to enter an evolution stage for searching for a non-dominant solution; otherwise, turning to the step C15; wherein/>Representing the maximum number of evolutions;
step C5: prediction using the application QoS prediction method described in step 2 And/>A target value for each solution;
Step C6: calling penalty functions And dominance function/>Ordered modification/>And/>Target value of invalid solution in (i.e./>)；
Step C7: calling a ranking functionPair/>And/>All solutions in (1) are ordered, i.e；
Step C8: calling a merge functionMerging/>And/>I.e.；
Step C9: using selection functionsSelecting a certain number of parent solutions from the merged population, i.e；
Step C10: creation of offspring populations using genetic operators；
Step C11: calling a merge functionOffspring population/>And/>Merging, i.e.；
Step C12: calling a merge functionOffspring population/>And/>Merging, i.e.；
Step C13: updating iteration counterThe formula is/>；
Step C14: invoking a selection functionUpdating population/>And/>The formula isStep C4, the back turning is carried out;
Step C15: after the iteration of the population with preset times, outputting the current population Terminating after a set of non-dominant solutions.
Penalty function in step C6Specifically, for population/>Each of which is invalid solution，/>Altering the corresponding target value/>I.e.Wherein/>Is an invalid solution/>Constraint violation of/>Representing generation-related parameters,/>Representing a maximum target value; for the dominance function/>Specifically, for a populationEach invalid solution/>，/>Altering the corresponding target value/>I.e.Wherein/>Is a weight vector associated with the search area.
In the architecture model used in the present invention, consider a modern medical edge device enabled by CPU-GPU collaborative MPSoC, integrationCPU big core/>、/>CPU corelet/>And a GPU core/>. Although current CPU-GPU co-MPSoC is equipped with multiple GPU cores, a general purpose setup is adopted, i.e., all GPU elements are abstracted to one core, because a particular GPU core cannot be explicitly specified to complete execution of a task. In this system, any two cores that are in the same cluster are considered to be homogeneous, while cores that are not in the same cluster are considered to be heterogeneous.
Furthermore, in the present invention, the cores of the same cluster all support DVFS techniques, which all operate at the same operating frequency, i.e., do not allow to increase or decrease the operating frequency of a given core while guaranteeing that the operating frequencies of other cores in the same cluster are unchanged. UsingRepresenting CPU big core/>Is a voltage frequency combination set of CPU corelet/>, in the same wayAnd GPU core/>Respectively denoted asAnd/>。
In the application model used in the present invention, use is made ofRepresenting periodic execution of a set of totals/>, on a target hardware platformIs provided. Scheduling order of these applications/>Predetermined by existing sorting methods. For each application/>Its workload/>Application program/>, as measured by number of execution cycles, is allowedAdjusting task workload/>Partition and mapping points between the CPU and GPU cores.Representing an array of candidate partition map points, where each element/>Representation/>The proportion of task workload on the CPU and GPU cores. No matter which partition and mapping point is adopted, the response delay of the application program is no later than the preset deadline/>Avoiding timing errors.
In the life reliability model used in the invention, three common wear faults are concerned, namely: time Dependent Dielectric Breakdown (TDDB), electromigration (EM), and Thermal Cycling (TC). TDDB refers to wear caused by the formation of conductive paths through a dielectric, expressed as，/>And/>Respectively a field acceleration factor and an electric field through the dielectric,/>Representing the temperature, parameters/>, of the coreIs a constant; EM is the current caused by the movement of metal atoms, and the failure rate caused by it is expressed as/>Wherein/>Is a constant,/>Is the current density; TC is the wear caused by thermal stress due to mismatch in the coefficients of thermal expansion of adjacent material layers, and the failure rate caused by this is expressed as/>，/>Is a constant related to the material layer,/>Is the Coffin-Manson index,/>Is the thermal cycle frequency,/>And/>Respectively the whole and the elastic part of the temperature cycle range. Based on the failure rate summation method, the life reliability of the core can be calculated as failure rate summation versus time/>The final life reliability model is:
in the QoS model used in the invention, the application program is caused to be Response delay running on target CPU-GPU collaborative MPSoC is/>Can be expressed as/>Wherein/>For application/>Start time of/>And/>Application program/>, respectivelyExecution time in the case of execution on CPU and GPU cores only. Similarly, let application/>Is/>Can be expressed asWherein/>And/>The power consumption of the CPU and GPU, respectively. To limit the QoS of an application, a preference function/>, is definedTo measure individual application preferences for response delay and power consumption, denoted/>Wherein/>Is a potential variable that is used to determine,Is a preference factor. At this point the computing application/>I.e./>The method comprises the following steps:
Wherein, And/>Is two decision variables.
The invention aims to accurately estimate the load of an application program by designing an application program load prediction method based on machine learning, develop a personalized drive application program QoS prediction method to correlate the application program QoS with response delay and power consumption preference of the application program, develop a group intelligent application program scheduling strategy based on c-DPEA by using two prediction methods, provide a personalized scheduling technology with ensured reliability for intelligent medical application programs energized by CPU-GPU cooperated with MPSoC, and maximize the QoS of the intelligent medical personalized application program on the premise of meeting all design constraints.
The objective function of the present invention can be expressed as:
Maximization:
constraint conditions:
Wherein, Representing application/>Upper response delay limit,/>Representing the overall power consumption of the application program,Representing energy budget,/>Representing peak temperatures of all cores,/>Representing the temperature threshold of all cores,/>Representing minimum lifetime reliability of all cores,/>Representing the reliability threshold for all cores.
A HARDKERNEL ODROID-XU3 board is adopted as a real hardware platform for realizing the QoS optimization solution, and the platform is developed based on Samsung Exynos 5422 MPSoC and integrates 4 ARM Cortex A15 cores, 4 ARM Cortex A7 cores and 1 ARM Mali-T628 MP6 GPU.4 ARM Cortex A15 cores form a high-performance CPU large core cluster, and each core supports a plurality of frequencies of 200 MHz-2000 MHz at intervals of 100MHz.4 ARM Cortex A7 cores form a low-power consumption CPU small core cluster, and each core supports a plurality of frequencies from 200MHz to 1400MHz at intervals of 100MHz. For ARM Mali-T628 MP6 GPU cores, the available frequencies can be selected from {600, 543, 480, 420, 350, 266, 177} MHz. In addition, we selected ten different classes of representative applications from the Hetero-Mark OpenCL suite, as shown in table 1, these measured benchmarks can cover a large number of MCPS application scenarios, facilitating a comprehensive comparison.
TABLE 1
The above was set as an experimental environment.
Comparing the present invention with eight other representative reference algorithms:
algorithm 1: SVR is a support vector regression algorithm, which is built on a mature statistical learning theory and shows satisfactory performance in solving the nonlinear regression problem.
Algorithm 2: LR is a traditional method of modeling the linear relationship between response and predicted variables, with the aim of learning a 0/1 classification model from training data features to deal with regression problems with dependent variables as classification variables.
Algorithm 3: NSGA-iii is a third generation non-dominant ranking genetic algorithm that uses a set of self-renewing reference points to maintain diversity among population members.
Algorithm 4: MOEA/D is a decomposition-based multi-objective evolutionary algorithm in which an initial problem is first split into multiple scalar optimization sub-problems, which are then solved in parallel to reduce computational complexity.
Algorithm 5: DLS-MOEA is an index-based algorithm for solving the multi-objective optimization problem. It developed a new solution generator integrated with a set of external profiles to direct the search direction to different sub-areas of the pareto front.
Algorithm 6: maOEA/IGD is a multi-objective optimization technique based on the back-off distance. In this approach, inverse algebraic distance (IGD) is used as an index to select an advantageous solution to maintain better algorithm convergence and population diversity.
Algorithm 7: MOALO is a multi-objective ant lion optimizer. Popular roulette methods are incorporated to guide promising ants (i.e., solutions) toward ideal search areas for multi-objective optimization problems.
Algorithm 8: TEEM is a hot aware application scheduling method for optimizing the average performance of all applications. However, it ignores other important constraints such as timing, energy budget, and life reliability.
Tables 2 and 3 list the accuracy of the workload prediction of the prediction method of the algorithm, SVR and LR benchmark algorithms of the present invention, and compare the improvement of the present invention with respect to the workload prediction of the SVR and LR algorithms. It can be seen that the present invention can still achieve higher prediction accuracy when the number of samples is relatively small. For example, when the number of samples is 100, the present invention improves the workload prediction accuracy of the application program KM by 5.8% and 23.6%, respectively, compared to the reference algorithms SVR and LR. It can also be seen from the table that all prediction methods can improve the accuracy of workload prediction when more samples are involved in the training process, but the method of the present invention is superior to the benchmark test algorithm.
TABLE 2
TABLE 3 Table 3
Fig. 2 depicts a confusion matrix used in the training process of the personality-driven application QoS prediction method of the present invention for evaluating the personality-driven application QoS prediction method. In this set of experiments, six applications KMeans, PB, ASE, BE, CH and BS were used to train the feature driven QoS prediction method of the present invention. The invention standardizes the QoS of each application program into an integer, the minimum value is 1, the maximum value is 10, and the larger the integer is, the higher the QoS realized by the application program is. In the confusion matrix of fig. 2, each diagonal element represents QoS prediction accuracy of a single application, while the remaining elements show prediction rates of output erroneous QoS scores. For example, when the QoS of the actual application is 4, the QoS prediction method of the present invention has a likelihood of successfully outputting the correct score of 81.3%. We observe that the prediction method of the present invention can maintain a prediction accuracy of between 76.5% and 87.0%. Fig. 3 investigates the generalization capability of the personality-driven QoS prediction method of the present invention for applications FDEB, FIR, KNN and EPs. As shown in fig. 3, the average and maximum prediction accuracy of our QoS prediction method is 87.8% and 93.5%, respectively. It can be seen that the personality-driven QoS prediction method developed by the present invention has a satisfactory generalization capability.
FIG. 4 is a graph showing the results of a comparison of the c-DPEA-based application scheduling method developed by the present invention with five typical non-dominant solution search algorithms NSGA-III, MOEA/D, DLS-MOEA, maOEA/IGD, and MoALO, for evaluating non-dominant solution search methods. In this set of comparison experiments, a metric called supersvolume was employed to jointly evaluate the convergence of the multi-objective optimization strategy and the diversity of the derivative solution set. A larger supersvolume is preferred here because it suggests that the corresponding algorithm will produce a non-dominant solution that is closer to the pareto boundary. As shown in FIG. 4, the c-DPEA based method of the present invention achieves significant capacity improvement when executing any of ten applications under test. Taking application ASE as an example, the superscales obtained by our c-DPEA based method and reference algorithms NSGA-III, MOEA/D, DLS-MOEA, maOEA/IGD and MOALO were 0.968, 0.773, 0.890, 0.867, 0.904 and 0.823, respectively.
Fig. 5 illustrates the application QoS achieved by the solution of the present invention and benchmarking scheme TEEM when running different applications. As can be seen from the figure, the solution of the present invention improves the application QoS by 15.7% on average. In addition, the method of the invention can obviously balance the QoS of a single application program and realize the effect of improving the balance by 64.3 percent.
Fig. 6 illustrates the scheduling feasibility of the solution and benchmarking scheme TEEM of the present invention. The scheduling feasibility of an algorithm refers to the ratio of the number of applied schedules to the number of all applied schedules (i.e., 1000 in the experiments of the present invention) that meets the specified constraints. For example, when considering only life reliability constraints in the equation, i.eThe scheduling feasibility of the solution of the present invention and baseline test scheme TEEM was 100% and 62%, respectively. When all design constraints are considered, the scheduling feasibility of the solution of the present invention is as high as 100%, 66.7% higher than the baseline solution TEEM. This is mainly because other important constraints such as timing, energy budget, and life reliability are ignored in benchmarking scheme TEEM. Thus, the results in FIG. 6 verify the effectiveness of our solution in enhancing the feasibility of personalized application scheduling.
From the above experimental data, it can be clearly seen that the present invention has good performance in both predicting workload accuracy and maximizing application QoS.
Example 2:
The embodiment of the invention also provides a personalized application scheduling device oriented to intelligent medical treatment, which comprises the following steps:
the acquisition module is used for acquiring the load of each application program;
The first processing module is used for obtaining the preference of the single intelligent medical application program for response delay and power consumption and QoS under the delay power consumption level according to the load of each application program;
And the second processing module is used for maximizing the QoS of the personalized intelligent medical application program according to the preference of the single intelligent medical application program for response delay and power consumption and the QoS under the delay power consumption level.
As one implementation of the embodiment of the present invention, the acquisition module predicts the load of each application program through machine learning.
As one implementation of the embodiment of the present invention, the first processing module predicts the preferences of individual smart medical applications for response delay and power consumption and QoS at multiple delay power consumption levels based on the personality-driven applications;
As one implementation of the embodiment of the invention, the second processing module maximizes QoS of the personalized smart medical application based on the c-DPEA population smart application scheduling policy.
The above embodiments are merely illustrative of the preferred embodiments of the present invention, and the scope of the present invention is not limited thereto, but various modifications and improvements made by those skilled in the art to which the present invention pertains are made without departing from the spirit of the present invention, and all modifications and improvements fall within the scope of the present invention as defined in the appended claims.
Claims (4)
1. The personalized application scheduling method for intelligent medical treatment is characterized by comprising the following steps of:
step S1, acquiring the load of each application program;
Step S2, obtaining preference of a single intelligent medical application program for response delay and power consumption and QoS under delay power consumption level according to the load of each application program;
step S3, maximizing the QoS of the personalized intelligent medical application program according to the preference of the single intelligent medical application program for response delay and power consumption and the QoS under the delay power consumption level;
in step S2, the preference of the individual smart medical application for response delay and power consumption and QoS at various delay power consumption levels are predicted based on the personality-driven application; the method specifically comprises the following steps:
Step B1: initializing loop flags ; Wherein/>The number of applications to participate in the questionnaire;
Step B2: initializing preference weights ；
Step B3: by means ofComputing application/>In the first placeResponse delay/>, at a single delay power consumption levelWherein/>Is/>Start time of/>Is/>At execution time on CPU core only,/>Is/>Execution time on GPU kernel only,/>Is/>Proportion of workload on CPU and GPU cores,/>，/>Is the total delay energy level;
Step B4: by means of Calculation/>In/>Power consumption at a delayed power consumption level/>Wherein/>Is the power consumption of CPU,/>Is the power consumption of the GPU;
Step B5: by means of Calculating latent variables/>I.e./>Preference for response delay and power consumption;
Step B6: by means of Calculation/>In/>QoS values at the individual delay energy levels/>Wherein/>And/>Is a decision variable;
Step B7: solving for Calculating decision variables/>And/>Optimum value/>And/>；
Step B8: will mark the bitThe value is 1, namely: /(I)；
Step B9: solving forCalculation/>Optimum value/>；
Step B10: comparing randomly generated preference weightsAnd calculated optimum/>Absolute difference/>；
Step B11: determining absolute differenceIf the value is smaller than the preset value, turning to the step B12; otherwise, turning to step B15;
Step B12: updating ，/>；
Step B13: updating，/>；
Step B14: judgingIf so, turning to the step B9; otherwise, turning to step B15;
Step B15: regenerating preference weights ；
Step B16: judgingIf so, turning to the step B3; otherwise, turning to step B17;
step B17: solving for Calculating a parameter matrix/>Optimum value/>; Wherein/>For preference weight matrix,/>； />For dimension/>Is a feature score matrix of (a);
Step B18: will mark the bit The value is 1, namely: /(I)；
Step B19: acquisition ofFeature score/>；
Step B20: by means ofCalculation/>At/>Delaying QoS at the power consumption level;
Step B21: by passing through Calculating the optimal value/>；
Step B22: updating，/>；
Step B23: judgingIf so, turning to the step B19; otherwise, go to step B24; wherein/>Is the total number of applications;
step B24: after application program preference and QoS are calculated, preference weight is returned Decision variable/>AndQoS for applications at delayed energy consumption levels, i.e./>；
In step S3, based on the c-DPEA group intelligent application program scheduling strategy, maximizing QoS of the personalized intelligent medical application program; the method specifically comprises the following steps:
Step C1: predicting an application set using the application load prediction method described in step S1 A workload of each application;
step C2: initializing an iteration counter I.e./>；
Step C3: random initialization of two populationsAnd/>；
Step C4: judgingIf so, turning to step C5 to enter an evolution stage for searching for a non-dominant solution; otherwise, turning to the step C15; wherein/>Representing the maximum number of evolutions;
step C5: prediction using the application QoS prediction method described in step 2 And/>A target value for each solution;
Step C6: calling penalty functions And dominance function/>Ordered modification/>And/>Target value of invalid solution in (i.e./>)；
Step C7: calling a ranking functionPair/>And/>All solutions in (1) are ordered, i.e；
Step C8: calling a merge functionMerging/>And/>I.e./>；
Step C9: using selection functionsSelecting a certain number of parent solutions from the merged population, i.e；
Step C10: creation of offspring populations using genetic operators；
Step C11: calling a merge functionOffspring population/>And/>Merging, i.e.；
Step C12: calling a merge functionOffspring population/>And/>Merging, i.e.；
Step C13: updating iteration counter，/>；
Step C14: invoking a selection functionUpdating population/>And/>，Step C4, the back turning is carried out;
Step C15: after the iteration of the population with preset times, outputting the current population Terminating after a set of non-dominant solutions.
2. The smart medical oriented personalized application scheduling method according to claim 1, wherein in step S1, a load of each application program is predicted based on machine learning.
3. A smart-medicine-oriented personalized application scheduling apparatus for implementing the smart-medicine-oriented personalized application scheduling method of any one of claims 1 to 2, comprising:
the acquisition module is used for acquiring the load of each application program;
The first processing module is used for obtaining the preference of the single intelligent medical application program for response delay and power consumption and QoS under the delay power consumption level according to the load of each application program;
a second processing module for maximizing QoS of the personalized smart medical application based on preferences of the individual smart medical application for response delay and power consumption and QoS at the delay power consumption level
Wherein the first processing module predicts preferences of individual smart medical applications for response delay and power consumption and QoS at multiple delay power consumption levels based on the personality-driven applications;
the second processing module maximizes QoS for the personalized smart medical application based on the c-DPEA community smart application scheduling policy.
4. The smart medical oriented personalized application dispatcher of claim 3, wherein the acquisition module predicts the load of each application by machine learning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310825753.5A CN116541154B (en) | 2023-07-07 | 2023-07-07 | Intelligent medical-oriented personalized application scheduling method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310825753.5A CN116541154B (en) | 2023-07-07 | 2023-07-07 | Intelligent medical-oriented personalized application scheduling method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116541154A CN116541154A (en) | 2023-08-04 |
| CN116541154B true CN116541154B (en) | 2024-05-03 |
Family
ID=87447516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310825753.5A Active CN116541154B (en) | 2023-07-07 | 2023-07-07 | Intelligent medical-oriented personalized application scheduling method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116541154B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110389820A (en) * | 2019-06-28 | 2019-10-29 | 浙江大学 | A kind of private clound method for scheduling task carrying out resources based on v-TGRU model |
| CN114444240A (en) * | 2022-01-28 | 2022-05-06 | 暨南大学 | Delay and service life optimization method for cyber-physical system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090210740A1 (en) * | 2008-02-14 | 2009-08-20 | Song Huang | Off-chip access workload characterization methodology for optimizing computing efficiency |
-
2023
- 2023-07-07 CN CN202310825753.5A patent/CN116541154B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110389820A (en) * | 2019-06-28 | 2019-10-29 | 浙江大学 | A kind of private clound method for scheduling task carrying out resources based on v-TGRU model |
| CN114444240A (en) * | 2022-01-28 | 2022-05-06 | 暨南大学 | Delay and service life optimization method for cyber-physical system |
Non-Patent Citations (3)
| Title |
|---|
| An Application-Level QoS Control Method Based on Local Bandwidth Scheduling;Yong Wang et al.;《Journal of Electrical and Computer Engineering》;20180606;第1-9页 * |
| CPU-GPU MPSoC 中使用寿命驱动的OpenCL 应用调度方法;曹坤 等;《计算机研究与发展》;第60卷(第5期);第976−991页 * |
| 云计算环境下虚拟机服务质量保证和评估方法: 研究综述;郝佳 等;《云南大学学报( 自然科学版) 》;第40卷(第6期);第1093-1107页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116541154A (en) | 2023-08-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhou et al. | Resource management for improving soft-error and lifetime reliability of real-time MPSoCs | |
| Arnaldo et al. | Multiple regression genetic programming | |
| Zhang et al. | Multi-objective particle swarm optimization approach for cost-based feature selection in classification | |
| Coskun et al. | Static and dynamic temperature-aware scheduling for multiprocessor SoCs | |
| Ïpek et al. | Efficiently exploring architectural design spaces via predictive modeling | |
| Huang et al. | Autonomous power management with double-Q reinforcement learning method | |
| Coskun et al. | Temperature management in multiprocessor SoCs using online learning | |
| Ipek et al. | Efficient architectural design space exploration via predictive modeling | |
| Huang et al. | A framework for scalable bilevel optimization: Identifying and utilizing the interactions between upper-level and lower-level variables | |
| Lee et al. | Efficiency trends and limits from comprehensive microarchitectural adaptivity | |
| Zhou et al. | Rethinking co-design of neural architectures and hardware accelerators | |
| Tang et al. | CPU–GPU utilization aware energy-efficient scheduling algorithm on heterogeneous computing systems | |
| Sayadi et al. | Scheduling multithreaded applications onto heterogeneous composite cores architecture | |
| Peng et al. | Energy‐Aware Scheduling of Workflow Using a Heuristic Method on Green Cloud | |
| Cao et al. | CPU-GPU cooperative QoS optimization of personalized digital healthcare using machine learning and swarm intelligence | |
| Sheikh et al. | Performance, energy, and temperature enabled task scheduling using evolutionary techniques | |
| Shamsa et al. | Concurrent application bias scheduling for energy efficiency of heterogeneous multi-core platforms | |
| Dey et al. | Deadpool: Performance deadline based frequency pooling and thermal management agent in dvfs enabled mpsocs | |
| CN113158435B (en) | Complex system simulation running time prediction method and device based on ensemble learning | |
| Bouzidi et al. | Map-and-conquer: Energy-efficient mapping of dynamic neural nets onto heterogeneous MPSoCs | |
| Shang et al. | A novel power management for CMP systems in data-intensive environment | |
| CN116541154B (en) | Intelligent medical-oriented personalized application scheduling method and device | |
| Butko et al. | Exploration of performance and energy trade-offs for heterogeneous multicore architectures | |
| Wang et al. | Enabling energy-efficient and reliable neural network via neuron-level voltage scaling | |
| Yang et al. | ReLeTA: reinforcement learning for thermal-aware task allocation on multicore |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |