CN117827420A - Embedded system-oriented parameter intelligent tuning method, server and system - Google Patents

Abstract

The invention discloses a parameter tuning method, a server and a system for an embedded system. The method comprises the following steps: receiving system information of an embedded system collected and sent by a client; the system information is generated according to a service program running in the embedded system; classifying the system information to obtain a classification result of the system information; processing the classification result by taking the performance index of the service program as a reference or taking the CPU energy consumption and the memory energy consumption of the embedded system as a reference to obtain a parameter tuning result of the distributed system; and sending the parameter tuning result to the client so that the client updates the parameters of the distributed system. The invention can adjust and optimize the parameters of the embedded system based on performance or energy consumption.

Description

Embedded system-oriented parameter intelligent tuning method, server and system

Technical Field

The invention belongs to the field of optimization of embedded systems, and particularly relates to a parameter tuning method, a server and a system for an embedded system.

Background

The embedded system is a control system platform capable of embedding equipment, generally comprises a software running environment and an operating system thereof, is suitable for various terminal control equipment, has single and stable software running, is a computer system designed by specific application, and has more single application service compared with the general computer system. Embedded systems currently have wide and deep applications in many fields, and the following is a brief introduction to the typical application field of embedded systems: (1) Smart home. The intelligent home uses the home as a platform, integrates facilities related to home life by utilizing an embedded system, such as a comprehensive wiring technology, a network communication technology, a security technology, an automatic control technology and an audio-video technology, builds an efficient management system for home facilities and family schedule matters, improves home safety, convenience, comfort and artistry, and realizes an environment-friendly and energy-saving living environment. (2) an industrial control system. The industrial control system is a requirement for large data volume and high-speed transmission of images, voice signals and the like, and is combined with the Ethernet and a control network. The networking wave of the industrial control system integrates various current popular technologies such as embedded technology, multi-standard industrial control network interconnection, wireless technology and the like, so that the development space of the industrial control field is expanded, and new development opportunities are brought. (3) automotive electronics. The automobile electronic is a generic term for a vehicle body automobile electronic control device and a vehicle-mounted automobile electronic control device. The electronic control device of the automobile body comprises an engine control system, a chassis control system and an electronic control system of the automobile body. The automotive electronics function is to improve the safety, comfort, economy and entertainment of the automobile. The embedded system is composed of sensors, microprocessors, executors, tens or hundreds of electronic components and parts thereof. (3) consumer electronics. Consumer electronics is an electronic product designed around consumer applications and closely related to life, work and entertainment, and finally, the purposes of freely selecting information and enjoying entertainment by consumers are achieved. Consumer electronics are mainly focused on electronic products purchased by individuals and intended for personal consumption. Such as digital products, bluetooth sound boxes, intelligent coffee machines and the like, provide low-quality and low-cost commodities for consumers by means of the characteristics of high efficiency, stability, economy and the like of an embedded system. (4) Internet of things. The internet of things is a component part of a new generation computer, namely the internet of things, except that the user end of the internet of things extends to any objects for information exchange and communication. The internet of things can be integrated with a computer through intelligent sensing and recognition technologies. The three key technologies of the Internet of things are as follows: sensor technology, RFID tags, embedded system technology. (5) security, medical treatment and logistics. The embedded system is also widely used in the related fields of security, medical treatment, logistics and the like, and performs various functions such as monitoring, management and the like under a complex environment by depending on the high efficiency, stability and profession of the embedded system.

Compared with the general system, the embedded system has the following characteristics: (1) lack of hardware resources. Hardware resources of embedded systems, such as CPU, memory, storage and bandwidth resources, are generally less than those of desktop systems or servers, and for example, the size of the storage space available for the desktop systems and the server systems is often more than 1T, but the storage space of many embedded systems, such as intelligent air conditioners, is only a few hundred K, i.e., the storage space of the embedded systems is only less than one ten thousandth of the former. (2) energy consumption is emphasized. Many embedded systems typically require 7X24 hours of operation, and thus are more energy intensive than desktop systems, such as intelligent access control, intelligent refrigerators, etc., all require time of day operation. In addition, many embedded devices such as portable cameras and intelligent toys require batteries or charging to provide energy, and the energy consumption directly determines the normal working time of the device, so that (3) the upper layer application scene is single, which is particularly important for the energy consumption. The embedded system generally meets a specific application scene design, such as a refrigerator, an access control, a toy car, health monitoring and the like, so that the workflow is generally single, and the task load is clearer.

Embedded systems involve a number of different classes of system parameters and are difficult to set and prone to error, mainly for the following reasons. (1) the parameter types are complex and numerous. If the parameters relate to different modules such as memory, CPU, threads/processes, network and the like, and the number of the parameters is large, the number of the parameters is more than 100 only for the memory related parameters of a Linux-based embedded system, and the ordinary skill is difficult to be familiar with all the parameters. (2) it is difficult to have a unified parameter optimization scheme. The application scenes of the embedded system are more diversified, and the parameter optimization scheme aiming at a certain scene is difficult to be suitable for all business scenes, so that the parameter optimization scheme aiming at different application scenes is generally needed.

Disclosure of Invention

Aiming at the problems that the parameter setting threshold of an embedded operation system is high and errors are easy to occur, the invention aims to provide a parameter tuning method, a server and a system for the embedded system, and the system can automatically tune the parameters of the embedded system based on performance or energy consumption.

The technical scheme of the invention comprises the following steps:

the parameter tuning method for the embedded system is characterized by being applied to a server, and comprises the following steps:

receiving system information of an embedded system collected and sent by a client; the system information is generated according to a service program running in the embedded system;

classifying the system information to obtain a classification result of the system information;

processing the classification result by taking the performance index of the service program as a reference or taking the CPU energy consumption and the memory energy consumption of the embedded system as a reference to obtain a parameter tuning result of the distributed system;

and sending the parameter tuning result to the client so that the client updates the parameters of the distributed system.

Further, the system information includes: CPU, memory, network, file IO, interrupt and process.

Further, the system information of the embedded system collected and sent by the client includes:

collecting system information of the embedded system based on an ebpf mechanism;

and sending the system information to a server.

Further, before the classifying the system information and obtaining the classification result of the system information, the method further includes:

and performing dimension reduction processing and denoising processing on the system information.

Further, the processing the classification result based on the CPU energy consumption and the memory energy consumption of the embedded system to obtain a parameter tuning result of the distributed system includes:

setting a processing delay threshold T (latency) of the event request;

setting a throughput rate threshold T (throughput) of the embedded system;

acquiring a processing delay R (latency) of an actual event request in the classification result and a throughput rate R (throughput) of the embedded system;

taking (latency) andR (throughput) > T (throughput) as constraint conditions, taking the minimum CPU energy consumption and the memory energy consumption of the embedded system as optimization targets, and processing CPU frequency modulation related parameters, memory related parameters and process/thread related parameters in the classification result by utilizing an optimal parameter optimization algorithm to obtain a parameter tuning result of the distributed system; wherein the parameter optimization algorithm comprises: genetic algorithms, gaussian process algorithms, or reinforcement learning algorithms.

Further, in the case that the parameter optimization algorithm is a genetic algorithm, the processing the CPU fm related parameter, the memory related parameter, and the process/thread related parameter in the classification result by using the optimal parameter optimization algorithm to obtain a parameter tuning result of the distributed system includes:

generating a plurality of parameter tuning tasks;

during initialization, each parameter tuning task randomly sets memory parameters, CPU frequency modulation parameters and process/thread parameters;

tuning each parameter tuning task respectively, and selecting an optimal tuning configuration scheme from the obtained tuning configuration schemes;

distributing the optimal tuning configuration scheme to each parameter tuning task, and carrying out the next round of parameter tuning tasks according to the genetic algorithm principle on the premise of meeting the parallelization mechanism based on local punishment;

and obtaining a parameter tuning result of the distributed system until the iteration is finished.

Further, the parameter tuning result includes: CPU frequency modulation related parameters, memory related parameters and process/thread related parameters; the CPU frequency modulation related parameters comprise: CPU working mode or working frequency related parameter; the memory related parameters include: setting the length of a cache queue and recovering a memory; the thread/thread related parameters mainly comprise the number of threads, the thread stack size setting and the thread priority.

The parameter tuning server for the embedded system is characterized by comprising:

the receiving module is used for receiving the system information of the embedded system collected and sent by a client; the system information is generated according to a service program running in the embedded system;

the classification module is used for classifying the system information and acquiring a classification result of the system information;

the tuning module is used for processing the classification result by taking the performance index of the service program as a reference or taking the CPU energy consumption and the memory energy consumption of the embedded system as a reference to obtain a parameter tuning result of the distributed system;

and the sending module is used for sending the parameter tuning result to the client so that the client updates the parameters of the distributed system.

An embedded system-oriented parameter tuning system, the system comprising:

an embedded system;

at least one client for:

collecting and transmitting system information of the embedded system; the system information is generated according to a service program running in the embedded system;

updating parameters of the distributed system based on the parameter tuning result;

at least one server for:

receiving the system information;

classifying the system information to obtain a classification result of the system information;

processing the classification result by taking the performance index of the service program as a reference or taking the CPU energy consumption and the memory energy consumption of the embedded system as a reference to obtain a parameter tuning result of the distributed system;

and sending the parameter tuning result to the client.

The parameter tuning method for the embedded system is characterized by being applied to a client, and comprises the following steps:

collecting system information of an embedded system;

the system information is sent to a server, so that the server classifies the system information, a classification result of the system information is obtained, and the classification result is processed by taking performance indexes of the service program as references or CPU energy consumption and memory energy consumption of the embedded system as references, so that a parameter tuning result of the distributed system is obtained, and then the parameter tuning result is sent to the client;

and updating the parameters of the distributed system based on the parameter tuning result.

Compared with the prior art, the invention has the following positive effects:

(1) Compared with the traditional manual tuning, the invention can automatically tune the parameters of the embedded system, and is more efficient;

(2) The invention provides an optimized objective function for energy consumption when the parameter is self-optimized, aiming at the characteristic that the embedded system pays attention to energy consumption; and meanwhile, optimizing policy migration during multi-objective optimization is supported.

(3) Compared with the existing automatic tuning tool, the system information acquisition method based on the ebpf occupies fewer client resources, is biased to an embedded service scene, and is more suitable for an embedded system;

(4) Compared with the existing automatic tuning tool, the invention explicitly provides optimization for parameters such as memory parameters (cache queue length setting, memory recovery parameters and the like), CPU frequency modulation parameters (CPU working mode or working frequency parameters and the like), process/thread parameters (thread number, thread stack size setting, thread priority and the like) and the like aiming at an embedded system.

(5) The invention supports two different concurrent tuning modes: step parallel and asynchronous parallel modes, and a parallelization mechanism based on local penalty is provided for more efficient acquisition of an optimal scheme.

Drawings

FIG. 1 is a flow chart of a method and a system for intelligently adjusting parameters of an embedded system.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is apparent that the embodiments described are merely specific embodiments of the present invention, and not all embodiments.

The invention discloses an intelligent parameter tuning method and system for an embedded system, which consists of a client and a server. The client is deployed in the embedded system, and the server can be deployed on the same machine as the client or on other machines. The client is mainly responsible for collecting embedded system information and updating system parameters, and the server is mainly used for storing data transmitted by the client, obtaining optimized parameters through iterative training and returning the parameters to the client.

The invention discloses an intelligent parameter tuning method and system for an embedded system, and the technical scheme is as follows.

The flow chart of the technical scheme is shown in fig. 1, and the whole flow comprises three modules: the system comprises an embedded system, a parameter self-tuning client and a parameter self-tuning server. The embedded system is a target system to be optimized, the client is responsible for interacting with the embedded system, and the server only interacts with the client.

Compared with a common server and a desktop system, the embedded system has special requirements, such as resource shortage, energy consumption and the like, and the application scene of the embedded system is more specific, and only one specific service, such as intelligent face recognition access control, intelligent sound and the like, is usually processed; while the services handled by desktop and server systems tend to be more complex, desktop systems typically handle email, office software, etc. simultaneously, and server systems typically handle web servers, database services, etc. simultaneously. Aiming at embedded characteristics, parameters optimized for an embedded system in the technical scheme are mainly concentrated on the following aspects: (1) The memory parameters mainly comprise length setting of a cache queue, memory reclamation and the like. The setting of the cache queue refers to that various data packets such as related protocol data packets of MQTT and the like are required to be processed in an embedded system, data is stored through the cache queue frequently in order to improve efficiency, at this time, parameters such as the size of a memory unit in the cache queue and the length of the queue have default values, if the setting is too large, memory space is wasted, if the setting is small, memory is required to be allocated frequently, and the caching effect is small. The memory reclamation refers to the release and reclamation of dirty pages or outdated memory pages, and the current memory reclamation strategy is to reclaim the dirty pages when the proportion of the dirty pages reaches a threshold value or the time reaches the threshold value, if the threshold value is set too large, the dirty pages occupy a large amount of memory resources, and if the threshold value is set too small, the memory is frequently refreshed, so that the performance is reduced. And (2) CPU frequency modulation optimization. The frequency modulation optimization of the CPU refers to saving energy consumption or improving efficiency by adjusting the working mode or the working frequency of the CPU, and as described above, the embedded system often performs a trade-off between performance and energy consumption, so that the tuning of the working mode or the working frequency of the CPU can help to achieve the purpose. The common selectable types of the CPU working modes include a performance priority mode, an energy consumption priority mode (energy saving mode) and the like, and the CPU working frequency is generally provided with a plurality of candidate values, so that the highest frequency and the lowest frequency of the CPU can be set. (3) thread-related parameter settings. The thread-related parameters mainly relate to the number of threads, thread stack size settings and thread priorities. The memory resources in the embedded system are usually limited, the size of the thread stack and the number of threads can obviously influence the occupation of the memory resources, the stack overflow error can be caused by the too small setting of the thread stack, and the memory is wasted due to the too large setting; setting the number of similar threads too small can cause tasks not to be processed in time to influence the system performance, and setting too large can cause resource waste. Thread priority refers to the priority of each thread when the system is scheduled, and threads with high priority can obtain more execution time.

The client side is used as one of two core modules of the intelligent tuning tool for the parameters of the embedded system, and mainly completes the following functions: (1) The system information collection based on ebpf includes CPU, memory, network, file operation, etc. The traditional system information is based on tools such as vmstat, and the like, and the scheme collects the system information based on the ebpf, so that the method has the advantages of low invasiveness, convenience, programmability and the like compared with the traditional mode ebpf scheme. The system information collected by the client is sent to the server for processing. (2) a parameter update mechanism. The parameter updating mechanism is to update the corresponding parameters on the embedded system after the client receives the optimized parameters transmitted by the server. And the parameters are updated through interfaces sysctl, procfs, sysfs and the like of Linux. Parameters under the directory/proc/sys/vm of the procfs interface relate to memory mechanism tuning:

the block_dump indicates whether to open the Block Debug mode for recording all read-write and Dirty Block write-back operations.

dirty_expires_centrisecs indicates that dirty data has been resident in memory beyond this value, and the pdflush process will write the data back to disk the next time. Default 3000, units are 1/100s.

dirty_ratio indicates that if the dirty data generated by a process reaches the percentage of the overall memory of the system, the process writes the dirty data back to disk by itself.

dirty_writeback_centrisecs indicates how long the pdflush process periodically intervals dirty data association disks in 1/100s.

The vfs_cache_pressure represents the tendency of the kernel to reclaim memory for directors and inodes; default value 100 indicates that the kernel will place directors and inode cache newspapers in a reasonable percentage according to pagecache and swapcaches; decreasing this value below 100 will result in cores tending to reserve directors and inode caches; above 100, cores will tend to reclaim directors and inode caches.

min_free_kbytes represents how much free memory (KB) the Linux VM is forced to reserve lowest.

nr_pdflush_threads represents the number of pdflush processes currently in progress, and in the case of high I/O load, the kernel will automatically increment more pdflush.

The override_memory specifies a policy of the kernel for memory allocation, which may be 0, 1, 2.0 means that the kernel will check if there is enough memory available for the application process. If sufficient, the memory application allows; otherwise, the memory application fails. 1 indicates that the kernel allows allocation of all physical memory regardless of the current memory state. And 2, allowing the kernel to allocate the memory which is searched by the sum of all physical memories and the exchange space.

The server side mainly comprises a data storage and intelligent parameter adjusting module, and the two modules are respectively described as follows: (1) The data storage module is used for mainly storing system information transmitted by the client and comprises information such as a CPU, a memory, a network and the like; (2) The intelligent tuning module supports tuning targeting performance and tuning targeting energy consumption simultaneously. The module comprises three sub-modules: statistical information analysis, recommendation model and parameter analysis sub-module. The statistical information analysis submodule mainly processes system information such as dimension reduction, denoising, classification and the like; the recommendation model mainly comprises a plurality of intelligent algorithms, such as a genetic algorithm, a Gaussian process, a reinforcement learning algorithm and the like, and parameters are optimized based on the algorithms; the parameter analysis submodule mainly processes parameters to be optimized, such as screening TopK parameters with the greatest influence on the current application scene. The intelligent parameter adjusting module pursues the embedded system to process the transaction in the shortest time when the performance is aimed, and pursues the energy consumption to ensure the transaction processing and simultaneously reduce the energy consumption as much as possible. The invention takes the performance index of the current application program as a benchmark when the performance is targeted, and takes the utilization rate of the CPU and the memory as a benchmark when the energy consumption is targeted. The traditional parameter tuning is mostly aimed at performance tuning, and the invention supports energy consumption tuning besides the performance tuning. The application scenario of the embedded system generally needs to consider the energy consumption problem (such as a battery-driven intelligent access control or an intelligent camera), so that the intelligent parameter adjusting module optimizes the energy consumption, and the objective function of energy consumption optimization is as follows:

Min(α×U _cpu +β×U _mem )

s.t.R(latency)<T(latency)

andR(throughput)>T(throughput)

where α+β=1, R (latency) and R (throughput) are the processing delay and throughput of the actual event request, and T (latency) and T (throughput) are the corresponding thresholds set by the user. As shown in the formula, the core scheme of energy consumption optimization is to reduce the utilization rate of the CPU and the memory as much as possible on the premise of meeting the requirements of user delay and throughput rate. On the basis of supporting performance tuning and energy consumption tuning, the invention supports the migration of a multi-objective optimization scheme: for the multi-objective optimization problem with multiple objectives, the present invention proposes to migrate knowledge about the multiple objectives individually: firstly, training a model for each target independently, then combining corresponding models of a plurality of targets (such as weighted average of a plurality of model parameters) to obtain a new model, and then performing searching optimization of a parameter space based on the new model.

The invention supports a multi-parallelization mode, simultaneously supports starting a plurality of tasks for tuning, and currently supports two modes: synchronous parallel and asynchronous parallel modes. Synchronous parallelism: each task selects one configuration from the recommended configurations for verification until all tasks complete the round of verification and then begin the next round of verification. Asynchronous parallelism: each task selects one configuration from the recommended configurations for verification, and for each task, the next round of verification is started immediately after the current verification is finished.

In order to avoid improving the parallelization efficiency, the invention provides a parallelization mechanism based on local punishment: the current task needs to be sufficiently distinct from the configuration schemes that other tasks are validating when sampling the optimal parameters to avoid parallelization of the configuration schemes that are validated for several tasks.

Taking a genetic algorithm as an example, the tuning principle of the invention is described below in a synchronous parallel mode: assuming that four concurrent tasks task1, task2, task3 and task4 are provided, when the four tasks task1, task2, task3 and task4 are initialized, parameters such as memory parameters (cache queue length setting, memory recovery parameters and the like), CPU frequency modulation parameters (CPU working mode or working frequency parameters and the like), process/thread parameters (thread number, thread stack size setting, thread priority and the like) and the like are randomly set, then when the four tasks task1, task2, task3 and task4 are finished, a configuration scheme with the best current tuning effect is selected, and the scheme of task1 is assumed, and because the core of a genetic algorithm is the scheme which is then optimized (namely, a few genes are mutated during inheritance) on the basis of the current optimal solution, the scheme of task1 is distributed to the task2, task3 and task4; on the premise that a parallelization mechanism based on local punishment is met, four tasks before the next round of tuning and optimization are started, according to the principle that most of parameter values are unchanged and a small part of parameter values are randomly changed in the principle of a genetic algorithm, small part of the current optimal scheme is randomly modified, and then the next round of iterative optimization is started until iteration is finished.

The parameter tuning method comprises the following specific steps:

(1) Running a service program in the embedded system, such as intelligent face recognition access control;

(2) The client starts to collect information of the embedded system, including information of a CPU, a memory, a network, a file IO, an interrupt, a process and the like, see step (1) of FIG. 1;

(3) The client sends the collected information to the server, see step (2) of fig. 1;

(4) After receiving the system information, the server side stores the information into a data storage module;

(5) The statistical information analysis sub-module reads the system information from the data storage module and processes the system information, such as dimension reduction, denoising and the like.

(6) The recommendation model submodule cooperates with the statistical analysis submodule and the parameter analysis submodule to regulate and optimize system parameters;

(7) After parameter tuning is finished, the server transmits the optimized parameters to the client, see step (3) of fig. 1;

(8) After receiving the optimized parameters of the server, the client updates the parameters of the embedded system, see step (4) of fig. 1, where the parameters of the embedded system tuning mainly include CPU fm related parameters, memory related parameters, and process/thread related parameters.

In an example, assuming that an embedded system is an intelligent sound box, the intelligent sound box often becomes an entrance of an intelligent home device, and an air conditioner, a refrigerator, a sweeping robot and other devices are controlled through the intelligent sound box, a protocol such as MQTT is usually required to be supported, and system parameters in the service scene are required to be optimized at present, and the implementation steps are as follows:

(1) The intelligent sound box normally operates and receives various household appliance operation instructions from the mobile phone app;

(2) The intelligent sound box sends an operation instruction to the household appliance;

(3) Executing an operation instruction by the household appliance and feeding back an operation result to the intelligent sound box;

(4) The intelligent sound box feeds back the received household appliance operation result to the mobile phone app;

(5) The client collects the system information such as the memory, CPU, process/thread, network and the like of the system in the steps (1), 2), 3 and 4;

(6) The client sends the collected system information to the server;

(7) The server side adjusts the parameters of the intelligent sound box system according to the received system information;

(8) The server returns the optimized parameters to the client;

(9) And the client receives the optimized parameters transmitted by the server and updates the parameters into the intelligent sound box system.

Although the specific embodiments of, and examples for, the invention are disclosed for illustrative purposes, the present invention is described in the following description in order to facilitate the understanding of the principles of the invention, it will be appreciated by those skilled in the art that: various alternatives, variations and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the invention should not be limited to the disclosure of the illustrated embodiment and the example drawings.

Claims (10)

1. The parameter tuning method for the embedded system is characterized by being applied to a server, and comprises the following steps:

receiving system information of an embedded system collected and sent by a client; the system information is generated according to a service program running in the embedded system;

classifying the system information to obtain a classification result of the system information;

processing the classification result by taking the performance index of the service program as a reference or taking the CPU energy consumption and the memory energy consumption of the embedded system as a reference to obtain a parameter tuning result of the distributed system;

and sending the parameter tuning result to the client so that the client updates the parameters of the distributed system.

2. The method of claim 1, wherein the system information comprises: CPU, memory, network, file IO, interrupt and process.

3. The method of claim 1, wherein the system information of the embedded system collected and transmitted by the client comprises:

collecting system information of the embedded system based on an ebpf mechanism;

and sending the system information to a server.

4. The method of claim 1, wherein the classifying the system information, before obtaining the classification result of the system information, further comprises:

and performing dimension reduction processing and denoising processing on the system information.

5. The method of claim 1, wherein the processing the classification result based on the CPU power consumption and the memory power consumption of the embedded system to obtain the parameter tuning result of the distributed system comprises:

setting a processing delay threshold T (latency) of the event request;

setting a throughput rate threshold T (throughput) of the embedded system;

acquiring a processing delay R (latency) of an actual event request in the classification result and a throughput rate R (throughput) of the embedded system;

taking (latency) andR (throughput) > T (throughput) as constraint conditions, taking the minimum CPU energy consumption and the memory energy consumption of the embedded system as optimization targets, and processing CPU frequency modulation related parameters, memory related parameters and process/thread related parameters in the classification result by utilizing an optimal parameter optimization algorithm to obtain a parameter tuning result of the distributed system; wherein the parameter optimization algorithm comprises: genetic algorithms, gaussian process algorithms, or reinforcement learning algorithms.

6. The method according to claim 5, wherein, in the case that the parameter optimization algorithm is a genetic algorithm, the processing the CPU fm related parameter, the memory related parameter, and the process/thread related parameter in the classification result by using the optimal parameter optimization algorithm to obtain the parameter tuning result of the distributed system includes:

generating a plurality of parameter tuning tasks;

during initialization, each parameter tuning task randomly sets memory parameters, CPU frequency modulation parameters and process/thread parameters;

tuning each parameter tuning task respectively, and selecting an optimal tuning configuration scheme from the obtained tuning configuration schemes;

distributing the optimal tuning configuration scheme to each parameter tuning task, and carrying out the next round of parameter tuning tasks according to the genetic algorithm principle on the premise of meeting the parallelization mechanism based on local punishment;

and obtaining a parameter tuning result of the distributed system until the iteration is finished.

7. The method of claim 1, wherein the parameter tuning results comprise: CPU frequency modulation related parameters, memory related parameters and process/thread related parameters; the CPU frequency modulation related parameters comprise: CPU working mode or working frequency related parameter; the memory related parameters include: setting the length of a cache queue and recovering a memory; the thread/thread related parameters mainly comprise the number of threads, the thread stack size setting and the thread priority.

8. The parameter tuning server for the embedded system is characterized by comprising:

the receiving module is used for receiving the system information of the embedded system collected and sent by a client; the system information is generated according to a service program running in the embedded system;

the classification module is used for classifying the system information and acquiring a classification result of the system information;

the tuning module is used for processing the classification result by taking the performance index of the service program as a reference or taking the CPU energy consumption and the memory energy consumption of the embedded system as a reference to obtain a parameter tuning result of the distributed system;

and the sending module is used for sending the parameter tuning result to the client so that the client updates the parameters of the distributed system.

9. An embedded system-oriented parameter tuning system, the system comprising:

an embedded system;

at least one client for:

collecting and transmitting system information of the embedded system; the system information is generated according to a service program running in the embedded system;

updating parameters of the distributed system based on the parameter tuning result;

at least one server for:

receiving the system information;

classifying the system information to obtain a classification result of the system information;

processing the classification result by taking the performance index of the service program as a reference or taking the CPU energy consumption and the memory energy consumption of the embedded system as a reference to obtain a parameter tuning result of the distributed system;

and sending the parameter tuning result to the client.

10. The parameter tuning method for the embedded system is characterized by being applied to a client, and comprises the following steps:

collecting system information of an embedded system;

the system information is sent to a server, so that the server classifies the system information, a classification result of the system information is obtained, and the classification result is processed by taking performance indexes of the service program as references or CPU energy consumption and memory energy consumption of the embedded system as references, so that a parameter tuning result of the distributed system is obtained, and then the parameter tuning result is sent to the client;

and updating the parameters of the distributed system based on the parameter tuning result.