CN111930484A - Method and system for optimizing performance of thread pool of power grid information communication server - Google Patents

Abstract

The invention discloses a method and system for optimizing the performance of a thread pool of a power grid information communication server. The method includes: analyzing factors affecting the performance of the thread pool, thereby establishing a performance model of the thread pool; In the thread pool tuning model of the machine, the hyperparameters of the trained thread pool tuning model are obtained; through the trained support vector machine prediction model, it is judged whether the current thread pool size is the best size, and if not, the thread pool is reset. , and select the thread pool feature data that meets certain conditions to dynamically update the training sample set; through the dynamic thread pool intelligent tuning model proposed in this solution, the user response time of the server can be intelligently reduced, especially during peak access times. to improve the performance of the server.

Description

A method and system for optimizing the performance of a thread pool of a power grid information communication server

technical field

The invention relates to the field of smart grid, in particular to a method and system for optimizing the performance of a thread pool of a power grid information communication server.

Background technique

With the development of my country's power grid towards intelligence, networking and automation, the information interaction between power information networks has become more frequent and in-depth. The power grid information communication server carries the core business of power grid information network information transmission, and often faces a large number of user requests, and the processing time required for these user tasks is generally very short. Therefore, the ICT server generally adopts the thread pool technology to respond to these user requests in a timely and efficient manner. However, while the thread pool improves system performance, it also raises a new problem, that is, how to choose an appropriate thread pool size to obtain the best server performance. If the size of the thread pool is too large, although the ability of the thread pool to process user task requests in parallel will increase, it will also increase the system overhead for maintaining such a large number of threads. It will inevitably lead to more intense competition for system resources, which is likely to cause the performance of the system to decline. If the size of the thread pool is too small, it will weaken the ability of the thread pool to process user requests in parallel. Therefore, choosing an appropriate thread pool size becomes a key factor in determining server performance.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a method and system for optimizing the performance of the thread pool of a power grid information communication server, which can select an appropriate thread pool size and reduce the user response time of the server.

The technical scheme that the present invention solves the above-mentioned technical problems is as follows:

A method for optimizing the performance of a power grid information communication server thread pool, comprising the following steps:

S1, analyze the factors that affect the performance of the thread pool, so as to establish a performance model of the thread pool;

S2, input the performance test data of the ICT server into the thread pool tuning model based on the support vector machine, and obtain the hyperparameters of the trained thread pool tuning model;

S3, judge whether the current thread pool size is the optimal size through the trained support vector machine prediction model, if not, reset the thread pool, and select the thread pool feature data that meets certain conditions to dynamically update the training sample set;

The thread pool tuning model is established according to the thread pool performance data throughput, task operation time, task blocking time and the corresponding optimal thread pool size, and the thread pool performance optimization is to select an appropriate thread pool according to the number of user requests size.

Further, the S1 specifically includes:

(1) Let the user task response time be t _response , the queue waiting time of the task in the queue is t _queue , and the task processing time in the pool in the thread pool is t _pool , then t _response = t _queue + t _pool ;

(2) The processing time of a task in the thread pool includes the task preempting CPU operation time t _operation and the waiting time t _{wait for} the task to be suspended due to waiting for system resources, that is, t _pool = t _operation + t _wait . Therefore, the end user task response time t _response = t _queue + t _operation + t _wait ;

(3) Suppose the system throughput is m, the thread pool size is n, and the task operation time is t _operation , then the mathematical model of task queuing time is t _queuing = f(n, m, t _pool ) = f(n, m, t _operation + t _wait );

(4) Assuming that the time consumed by waiting for system resources is blocked as T _blocking , and the time that threads in the pool occupy for CPU operations is T _operation , the mathematical model of task waiting time can be written as t _waiting = g(n, T _operation , T _blocking );

(5) The task operation time t _operation refers to the time consumed by the user task to preempt the CPU to execute the task after entering the thread pool. For each user task, its operation time can be considered as a constant, independent of other parameters such as throughput and thread pool size, t _operation = T _operation ;

(6) To sum up, the mathematical model of user response time reflecting thread pool performance can be built as

_tresponse = _{tqueue+toperation + twait}

=f(n,m,T _operation +g(n,T _operation ,T _blocking ))

+T _operation +g(n, T _operation , T _blocking ),

It can be written as t _response = h(n, m, T _operation , T _blocking );

(7) To optimize the performance of the thread pool, that is, to minimize the user task response time t _response . If the above formula is continuously differentiable, the necessary condition for obtaining the minimum value is _t'response =h'(n _best , m, T _operation , T _blocking )=0.

Further, the S2 specifically includes:

S21, based on an improved fluid search optimization algorithm (IFSO), initialize the hyperparameters of the support vector machine; wherein, the hyperparameters include: a penalty factor C, a parameter γ of a radial basis kernel function;

S22 , cross-training is performed using a support vector machine, and an iterative optimization is performed according to the obtained classification accuracy as the fitness function of the IFSO, and the optimal hyperparameters are finally obtained.

Further, the S22 specifically includes:

(1) Initialize the position and velocity of each fluid particle, the density of the fluid, the direction of motion, and the normal pressure;

(2) Calculate the objective function value, update the optimal objective function value, the optimal position and the worst objective function value, and calculate the fluid particle density;

(3) Normalize the objective function value and calculate the pressure of the fluid particles;

(4) Calculate the pressure and velocity direction of other fluid particles to the current particle;

(5) Calculate fluid velocity value and velocity vector according to Bernoulli equation;

(6) Update the position of the particle

(7) Repeat steps (2)-(6) until the termination condition is satisfied.

Among them, in order to improve the accuracy of the fluid search algorithm, a two-stage optimization mechanism is adopted, namely the diversified search in the first stage and the refined exploration in the second stage.

Further, the S3 specifically includes:

Input the real-time running thread pool performance monitoring data as a test sample into the support vector machine to obtain the best thread pool size category to which it belongs;

Determine whether the obtained optimal thread pool size is consistent with the current size, if not, reset the thread pool and dynamically adjust the thread pool size

Determine whether the feature data satisfies the KKT (Karush-Kuhn-Tucker) condition, if so, replace the point that most violates the KKT condition in the training sample set, train and learn through the support vector machine, and get the new classification super of each optimal thread pool size. flat.

The beneficial effect of adopting the above-mentioned further scheme is: in the thread pool tuning model based on support vector machine, the KKT condition is used as the judgment condition for the update of the training sample set. The continuous introduction of samples makes the sample set increase infinitely, which can make the thread pool tuning model adapt to the complex and changeable environment.

A system for optimizing the performance of a thread pool of a power grid information communication server, comprising: a function relationship establishing module with optimal thread pool performance, a support vector machine parameter selection module and a thread pool size tuning module;

The function relationship establishing module with the optimal thread pool performance is used to analyze the factors affecting the thread pool performance, so as to optimize the thread pool performance, the purpose of optimizing the server performance can be achieved;

The support vector machine parameter selection module is used to input the performance test data of the communication server into the thread pool tuning model based on the support vector machine, and obtain the hyperparameters of the trained thread pool tuning model;

The thread pool size tuning module is used to judge whether the current thread pool size is the optimal size through the trained support vector machine prediction model, if not, reset the thread pool, and select the thread pool feature data dynamic that meets certain conditions. Update the training sample set;

The thread pool tuning model is established according to the thread pool performance data throughput, task operation time, task blocking time and the corresponding optimal thread pool size, and the thread pool performance optimization is to select an appropriate thread pool according to the number of user requests size.

Further, the function relationship establishment module with the best performance of the thread pool is used to analyze the factors affecting the performance of the thread pool, and the steps specifically include:

(1) Let the user task response time be t _response , the queue waiting time of the task in the queue is t _queue , and the task processing time in the pool in the thread pool is t _pool , then t _response = t _queue + t _pool ;

(2) The processing time of a task in the thread pool includes the task preempting CPU operation time t _operation and the waiting time t _{wait for} the task to be suspended due to waiting for system resources, that is, t _pool = t _operation + t _wait . Therefore, the end user task response time t _response = t _queue + t _operation + t _wait ;

(3) Suppose the system throughput is m, the thread pool size is n, and the task operation time is t _operation , then the mathematical model of task queuing time is t _queuing = f(n, m, t _pool ) = f(n, m, t _operation + t _wait );

(4) Assuming that the time consumed by waiting for system resources is blocked as T _blocking , and the time that threads in the pool occupy for CPU operations is T _operation , the mathematical model of task waiting time can be written as t _waiting = g(n, T _operation , T _blocking );

(5) The task operation time t _operation refers to the time consumed by the user task to preempt the CPU to execute the task after entering the thread pool. For each user task, its operation time can be considered as a constant, independent of other parameters such as throughput and thread pool size, t _operation = T _operation ;

(6) To sum up, the mathematical model of user response time reflecting thread pool performance can be built as

_tresponse = _{tqueue+toperation + twait}

=f(n,m,T _operation +g(n,T _operation ,T _blocking ))

+T _operation +g(n, T _operation , T _blocking ),

It can be written as t _response = h(n, m, T _operation , T _blocking );

(7) To optimize the performance of the thread pool, that is, to minimize the user task response time t _response . If the above formula is continuously differentiable, the necessary condition for obtaining the minimum value is _t'response =h'(n _best , m, T _operation , T _blocking )=0.

Further, the SVM parameter selection module includes: a SVM parameter initialization module and a SVM parameter training module;

The support vector machine parameter initialization module is used to initialize the hyperparameters of the support vector machine; wherein, the hyperparameters include: a penalty factor C, a parameter γ of the radial basis kernel function;

The support vector machine training module is used for iterative optimization according to the obtained classification accuracy as the fitness function of the IFSO, and finally the optimal hyperparameters are obtained.

Further, the support vector machine parameter training module is specifically used to calculate the optimal hyperparameters suitable for the thread pool size tuning module, and the steps specifically include:

(1) Initialize the position and velocity of each fluid particle, the density of the fluid, the direction of motion, and the normal pressure;

(2) Calculate the objective function value, update the optimal objective function value, the optimal position and the worst objective function value, and calculate the fluid particle density;

(3) Normalize the objective function value and calculate the pressure of the fluid particles;

(4) Calculate the pressure and velocity direction of other fluid particles to the current particle;

(5) Calculate fluid velocity value and velocity vector according to Bernoulli equation;

(6) Update the position of the particle;

(7) Repeat steps (2)-(6) until the termination condition is satisfied.

Among them, in order to improve the accuracy of the fluid search algorithm, a two-stage optimization mechanism is adopted, namely the diversified search in the first stage and the refined exploration in the second stage.

Further, the thread pool size tuning module is specifically used for:

Input the real-time running thread pool performance monitoring data as a test sample into the support vector machine to obtain the best thread pool size category to which it belongs;

Determine whether the obtained optimal thread pool size is consistent with the current size, if not, reset the thread pool and dynamically adjust the thread pool size

Judge whether the feature data satisfies the KKT (Karush-Kuhn-Tucker) condition, and if so, replace the point that most violates the KKT condition in the training sample set, and pass it to the support vector machine training module to generate a new classification of the optimal thread pool size. flat.

The beneficial effects of adopting the above-mentioned further scheme are:

The beneficial effects of the present invention are:

The present invention constructs the original training sample set by using a large number of communication server performance test data, dynamically adjusts the size of the thread pool under different power grid scenarios through the trained support vector machine, and uses the real-time performance test data to conduct the training sample set. Dynamic update can realize dynamic and intelligent tuning of the ICT server.

In the thread pool tuning model based on support vector machine, the KKT condition is used as the judgment condition for the update of the training sample set. For the training sample set, a fixed size is maintained to avoid the infinite sample set with the continuous introduction of new samples. If it increases, the thread pool tuning model can adapt to complex and changeable environments.

Advantages of additional aspects of the invention will be set forth in part in the description that follows, and parts will be apparent from the description below, or learned by practice of the invention.

Description of drawings

1 is a schematic flowchart of a method for optimizing the performance of a power grid information communication server thread pool according to an embodiment of the present invention;

2 is a flowchart of an IFSO-based support vector machine algorithm provided by other embodiments of the present invention;

3 is a normalized partial training sample set data provided by an embodiment of the present invention;

Fig. 4 is the classification accuracy rate that the IFSO-SVM provided by the embodiment of the present invention compares different algorithms to optimize the SVM obtained;

5 is an iterative curve diagram of the classification accuracy obtained by IFSO-SVM compared to different algorithm optimization SVMs according to an embodiment of the present invention;

6 is a comparison result of the performance of the dynamic thread pool and the static thread pool intelligently adjusted by different optimization algorithms provided by an embodiment of the present invention;

FIG. 7 is an efficiency improvement result of IFSO-SVM provided by an embodiment of the present invention compared with different algorithms;

FIG. 8 is a structural diagram of a system for optimizing the performance of a thread pool of a power grid information communication server according to an embodiment of the present invention.

Detailed ways

The principles and features of the present invention will be described below with reference to the accompanying drawings. The embodiments are only used to explain the present invention, but not to limit the scope of the present invention.

In the power communication network environment, the ICT server generally uses thread pool technology to respond to these user requests in a timely and efficient manner. The server creates a set of threads for network applications in advance. When a user service request arrives, it can directly call the thread pool to create a set of threads. A good thread serves the user. When the user task is completed, this group of threads is not destroyed, but waits for the service request of the next batch of users. However, while the thread pool improves system performance, it also creates a new problem. If the size of the thread pool is too large, although it will increase the thread pool's ability to process user task requests in parallel, it will also increase the system's ability to maintain such a large number of tasks. If the size of the thread pool is too small, it will weaken the ability of the thread pool to process user requests in parallel. The thread pool performance optimization problem refers to selecting an appropriate thread pool size according to the number of user requests.

As shown in FIG. 1, a method for optimizing the performance of a power grid information communication server thread pool provided by an embodiment of the present invention includes:

S1, analyze the factors that affect the performance of the thread pool, so as to establish a performance model of the thread pool;

(1) Let the user task response time be t _response , the queue waiting time of the task in the queue is t _queue , and the task processing time in the pool in the thread pool is t _pool , then t _response = t _queue + t _pool ;

(2) The processing time of a task in the thread pool includes the task preempting CPU operation time t _operation and the waiting time t _{wait for} the task to be suspended due to waiting for system resources, that is, t _pool = t _operation + t _wait . Therefore, the end user task response time t _response = t _queue + t _operation + t _wait ;

(3) Suppose the system throughput is m, the thread pool size is n, and the task operation time is t _operation , then the mathematical model of task queuing time is t _queuing = f(n, m, t _pool ) = f(n, m, t _operation + t _wait );

(4) Assuming that the time consumed by waiting for system resources is blocked as T _blocking , and the time that threads in the pool occupy for CPU operations is T _operation , the mathematical model of task waiting time can be written as t _waiting = g(n, T _operation , T _blocking );

(5) The task operation time t _operation refers to the time consumed by the user task to preempt the CPU to execute the task after entering the thread pool. For each user task, its operation time can be considered as a constant, independent of other parameters such as throughput and thread pool size, t _operation = T _operation ;

(6) To sum up, the mathematical model of user response time reflecting thread pool performance can be built as

_tresponse = _{tqueue+toperation + twait}

=f(n,m,T _operation +g(n,T _operation ,T _blocking ))

+T _operation +g(n, T _operation , T _blocking ),

It can be written as t _response = h(n, m, T _operation , T _blocking );

(7) To optimize the performance of the thread pool, that is, to minimize the user task response time t _response . If the above formula is continuously differentiable, the necessary condition for obtaining the minimum value is _t'response =h'(n _best , m, T _operation , T _blocking )=0.

Therefore, it can be concluded that the factors affecting the performance of the thread pool are throughput, task operation time, task blocking time and thread pool size. For the actual situation, the throughput, task operation time and task blocking time belong to the user task part and cannot be adjusted forcibly. However, the thread pool size can be dynamically adjusted in the thread pool program to adapt to objective user task changes and achieve the purpose of thread pool tuning;

S2, input the performance test data of the ICT server into the thread pool tuning model based on the support vector machine, and obtain the hyperparameters of the trained thread pool tuning model;

In a certain embodiment, a thread pool tuning model based on a support vector machine is established according to the thread pool performance data throughput, task operation time, task blocking time and the corresponding optimal thread pool size.

Support vector machine (SVM) is an intelligent method with strong generalization ability. The principle of the support vector machine is to set the input training sample points (x _i , y _i ), i=1...n, x∈R ⁿ , y∈{-1,+1} The hyperplane equation separated by (w·x )+b=0. There can be an infinite number of hyperplane equations, and the optimal hyperplane can be solved by the following formula:

where C is the penalty factor and ξ is the slack variable.

Introducing the Lagrangian factor α _i and transforming it into a dual problem, we get

i＝1,2,…,n且The optimal solution is

i=1,2,...,n and

So the optimal hyperplane equation is

Among them, SV is the support vector set, and the summation in the above formula is actually only performed on the support vector. Thus, the unknown samples can be classified by judging the value of y.

For the nonlinear case, it can be transformed into a linear partition problem of high-dimensional feature space by introducing the kernel function K(x·y)=φ(x)·φ(y), and a nonlinear support vector machine can be obtained. The dual problem is

Therefore, only the solution of this problem is required to obtain the optimal classification hyperplane. The decision function of the support vector machine is:

Commonly used kernel functions include linear kernel functions, polynomial kernel functions and radial basis kernel functions. Because the radial basis kernel function has the ability to map nonlinear classification problems into linear classification problems in infinite dimensional space, it has been widely used in the research of support vector machines. The radial basis kernel function is defined as:

K(x _i ,x)=exp(-γ||x _i -x|| ² ),γ>0

Among them, γ is the parameter of the radial basis kernel function.

The prediction performance of the support vector machine is determined by the penalty factor C and the parameter γ of the radial basis kernel function. Therefore, how to properly select the values of C and γ so that the prediction performance of SVM can be optimized has become a problem that needs to be focused on. The more direct method is to use the exhaustive method, but the computational load is large, and the optimal parameter combination may not be able to be searched. More methods are to use swarm intelligence optimization algorithm to search for the optimal parameter combination. In a certain embodiment, a relatively new and better-performing fluid search optimization algorithm is introduced and improved in order to improve the prediction performance of the support vector machine.

The basic steps of the FSO algorithm are as follows:

(1) Initialize the position of each fluid particle and adjust the parameters. Initialize fluid particle velocity _Vi = 0, fluid motion direction direction = 0, fluid density ρ _i =1, and normal pressure p ₀ =1 (i=1, 2...n).

(2) Calculate the objective function value, update the optimal objective function value y _best , the optimal position X _best and the worst objective function value y _worst . Calculate the fluid particle density ρ=m/l ^D .

(3) Normalize the objective function value and calculate the pressure of the fluid particles

(4) Calculate the pressure of other fluid particles on the current particle,

The speed direction is

并由步骤(4)中的方向计算出速度矢量V_i＝direction·v_i·rand(5) Calculate the fluid velocity value according to the Bernoulli equation

And calculate the velocity vector V _i =direction·vi·rand from the direction in step (4 ₎

(6) Location update. X _i+1 =X _i +V _i .

(7) Repeat steps (2)-(6) until the termination condition is satisfied.

Since the original FSO algorithm has a large amount of computation and is easy to fall into local optimum, this paper makes the following two improvements according to the actual situation of support vector machine optimization:

Improvement 1. Due to the large amount of calculation in the pressure direction in step (4), this paper simplifies it as

Improvement 2. In order to improve the accuracy of the fluid search algorithm, a two-stage optimization mechanism is adopted: the first-stage diversified search and the second-stage refined exploration. When the number of iterations reaches a certain threshold M' at the end of the first stage, the search space is shrunk to around the current optimal value, and the pixel length is reduced exponentially, i.e. l=l·e ^{(M'-t)/ σ} for refined exploration, where σ can set the precision of the search.

Figure 2 shows the support vector machine algorithm based on IFSO. Firstly, the values of hyperparameters C and γ of SVM are randomly initialized by IFSO, and then assigned to SVM for cross-training.

The thread pool performance data throughput, task computing time, task blocking time and optimal thread pool size obtained through experiments constitute a sample in the training sample set. Among them, throughput, task operation time, and task blocking time constitute the three characteristic attributes in the sample, and the optimal thread pool size constitutes the classification label of the sample. Through the learning and training of the support vector machine, the corresponding relationship between the feature attribute and the classification label is obtained. In this way, when the new feature data, that is, the test sample, arrives, the classification label corresponding to the test sample, that is, the optimal thread pool size, can be obtained according to the corresponding relationship, so as to provide a basis for adjusting the size of the thread pool. The effect of thread pool tuning is directly related to the selection of the SVM training sample set. The more comprehensive the training samples are, the more accurate the optimal thread pool size will be. Figure 1 shows the framework of the thread pool tuning model based on support vector machine, in which the specific steps of the training process of the thread pool tuning model based on support vector machine are:

Step1: First, classify the thread pool performance characteristic data through experiments, take throughput, task operation time and task blocking time as characteristic variables, divide them into multiple categories according to the optimal thread pool size, and construct an initial training sample set;

Step 2. Train and learn through the support vector machine, and obtain the classification hyperplane of each optimal thread pool size.

S3: Determine whether the current thread pool size is the optimal size through the trained support vector machine prediction model, if not, reset the thread pool, and select the thread pool feature data that meets certain conditions to dynamically update the training sample set.

The effect of thread pool tuning is directly related to the selection of the SVM training sample set. The more comprehensive the training samples are, the more accurate the optimal thread pool size will be. At the same time, due to the complexity and change of the actual situation, the fixed training sample set selected in advance should also be changed in order to adapt to the emergence of new and changing situations. To this end, the thread pool feature data obtained by monitoring is judged by certain conditions, and those that meet the conditions are used to dynamically update the training sample set, and those that do not meet the conditions are directly discarded. The specific steps are:

Step 1 Input the real-time running thread pool performance monitoring data as a test sample into the support vector machine, and obtain the best thread pool size category to which it belongs.

Step2: Determine whether the obtained optimal thread pool size is consistent with the current size. If not, reset the thread pool and dynamically adjust the thread pool size.

Step 3 judges whether the feature data satisfies the KKT (Karush-Kuhn-Tucker) condition, if so, replace the point that most violates the KKT condition in the training sample set, and return to Step 2; if it violates, discard and ignore.

In the thread pool tuning model based on support vector machine, this paper takes the KKT condition as the judgment condition for the update of the training sample set. If the test sample satisfies the KKT condition, it means that the sample will also fall into the area of the support vector, which will contribute to the classification decision function. It is necessary to update the training sample set and retrain the support vector machine; otherwise, if the test sample does not meet the KKT condition, It will not fall into the support vector region, and will not work for the regression decision function, so there is no need to retrain the support vector machine.

For the training sample set, a fixed size should be maintained, and the sample set should not grow infinitely with the continuous introduction of new samples. Therefore, while introducing new samples, it is necessary to remove the samples that play the least effect on the classification decision function, that is, the samples that violate the KKT condition most seriously, so that the number of training sample sets can be kept unchanged.

The original training sample set is constructed from a large number of ICT server performance experimental data, and the size of the thread pool in different power grid scenarios is dynamically adjusted by the trained support vector machine, and the training sample set is dynamically updated with real-time performance experimental data. , which can realize dynamic and intelligent tuning of the ICT server.

Preferably, in any of the above embodiments, S2 specifically includes:

S21, based on an improved fluid search optimization algorithm (IFSO), initialize the hyperparameters of the support vector machine; wherein, the hyperparameters include: a penalty factor C, a parameter γ of a radial basis kernel function;

S22 , cross-training is performed using a support vector machine, and an iterative optimization is performed according to the obtained classification accuracy as the fitness function of the IFSO, and the optimal hyperparameters are finally obtained.

By adopting the thread pool tuning model based on IFSO-SVM, higher classification accuracy can be obtained and the optimization effect of FSO can be improved, making it easier to jump out of the local optimum, and intelligently reducing the user response time of the server, especially during peak access times. It can play the role of peak shaving, which can improve the execution efficiency of the server.

Preferably, in any of the above embodiments, S22 specifically includes:

(1) Initialize the position and velocity of each fluid particle, the density of the fluid, the direction of motion, and the normal pressure;

(2) Calculate the objective function value, update the optimal objective function value, the optimal position and the worst objective function value, and calculate the fluid particle density;

(3) Normalize the objective function value and calculate the pressure of the fluid particles;

(4) Calculate the pressure and velocity direction of other fluid particles to the current particle;

(5) Calculate fluid velocity value and velocity vector according to Bernoulli equation;

(6) Update the position of the particle;

(7) Repeat steps (2)-(6) until the termination condition is satisfied.

Among them, in order to improve the accuracy of the fluid search algorithm, a two-stage optimization mechanism is adopted, namely the diversified search in the first stage and the refined exploration in the second stage.

By simplifying the calculation method of the pressure direction, and dividing the optimization process into two stages: diversified search and refined exploration, it can avoid the shortcomings of the original FSO often falling into local optimum and the large amount of calculation, thus greatly improving the model training. the speed of convergence.

Preferably, in any of the above embodiments, S3 specifically includes:

Input the real-time running thread pool performance monitoring data as a test sample into the support vector machine to obtain the best thread pool size category to which it belongs;

Determine whether the obtained optimal thread pool size is consistent with the current size, if not, reset the thread pool and dynamically adjust the thread pool size

Determine whether the feature data satisfies the KKT (Karush-Kuhn-Tucker) condition, if so, replace the point that most violates the KKT condition in the training sample set, train and learn through the support vector machine, and get the new classification super of each optimal thread pool size. flat.

In the thread pool tuning model based on support vector machine, by using the KKT condition as the judgment condition for the update of the training sample set, maintaining a fixed size for the training sample set can avoid the continuous introduction of new samples. The set grows infinitely and enables the thread pool tuning model to adapt to complex and changing environments.

In other embodiments provided by the present invention, according to the thread pool performance optimization method of this solution, a certain power grid information communication server (main configuration of the server: CPU is 2.4GHz Intel Xeon E5-2665, memory is 32G, and hard disk is 10T.) The throughput, task operation time and task blocking time collected in real time are characteristic variables, and the size of the thread pool is intelligently adjusted by the support vector machine. The data collection interval is to collect a set of server characteristic data every 15 minutes, and a total of 160 sets of data are collected during one week's working day. 160 sets of characteristic data are simulated through simulation experiments, and the user task response time is used as the evaluation index of server performance, and the optimal thread pool size is determined to construct a training sample set.

At the same time, in order to eliminate the dimensional influence between the features of the collected data and facilitate the training of SVM, the feature data is normalized.

Among them, d is the original value of the feature data, _dmin is the minimum value in the feature data, _dmax is the maximum value of the feature data, and g is the normalized feature data value. The normalized part of the training sample set data is shown in Figure 3.

In order to verify the performance of IFSO-SVM, the experiments compared the experimental results of the original fluid search algorithm FSO-SVM, particle swarm optimization algorithm PSO-SVM and artificial bee colony algorithm ABC-SVM. The support vector machine classifier adopts the open source software LIBSVM, and the average classification accuracy of the model is evaluated by 5-fold cross-validation. The parameters of each algorithm are set as follows: the number of particles is set to 30, and the maximum number of iterations is 50. The parameters of FSO and IFSO are set as follows: density limit ratio θ=20%, diversified search ratio M'=0.7, σ=40; particle swarm parameters are set as follows: c1=c2=2, ω _begin =0.9, ω _end =0.2 ; The parameters of artificial bee colony algorithm are P _onlooker =P _employed = 0.5; firefly algorithm α = 0.5, β _min = 0.2, γ = 1. The search range of parameter C in SVM is [0.01, 35000], and the search range of γ is [0.0001, 32].

Experiment 1: IFSO-SVM classification result test. As can be seen from Figure 4, IFSO-SVM achieves the highest classification accuracy among all contrasting algorithms. It is followed by ABC-SVM, FSO-SVM, FA-SVM and PSO-SVM. The average classification accuracy of the improved IFSO-SVM is 1.82 percentage points higher than the original FSO-SVM, and 1.61 percentage points higher than the second-placed ABC-SVM. It can be seen that the improvement of FSO makes it easier to search for the global optimal solution and improves the classification accuracy of SVM.

Figure 5 shows the iterative curves of the classification accuracy obtained by optimizing the SVM with different algorithms. It can be seen that at the beginning of the iteration, since the SVM parameters are randomly initialized by each algorithm, the obtained classification accuracy is basically the same, which is about 85%. As the search iteration proceeds, IFSO-SVM has been continuously improved, and finally obtained a higher classification accuracy than other optimization algorithms, significantly higher than the results of ABC-SVM and FSO-SVM below. It is worth noting that the original FSO-SVM performs slightly worse than the ABC-SVM, while the IFSO outperforms the original FSO in the classification accuracy in the iteration process, indicating that the improvement of IFSO enhances the search effect of FSO and makes it easier Jumping out of the local optimum can get a better classification effect.

Experiment 2: Server performance test. The performance test results of a certain burst of access to the power grid communication server, Figure 6 shows the performance comparison results of the dynamic thread pool and static thread pool (the size of the thread pool is fixed at 30) intelligently adjusted by different optimization algorithms. The pool adjustment interval is 1 minute. The whole experiment lasted for 45 minutes, of which 11 minutes to 33 minutes were the peak period of traffic. It can be clearly seen from Figure 4 that the performance of intelligent dynamic pools based on various optimization algorithms is significantly better than that of static pools. In the dynamic pool comparison of various optimization algorithms, the test results of the original FSO-SVM and ABC-SVM are comparable, and the results are similar to those of SVM training. The user response time based on IFSO-SVM dynamic pool is less than other comparison algorithms. Especially during the peak traffic period, the user response time of the IFSO-SVM dynamic pool is obviously more gradual than other optimization algorithms, which shows that IFSO has a good effect on the improvement of the original FSO, and can intelligently adjust the server response time. The role of peak clipping.

Figure 7 shows the efficiency improvement results of IFSO-SVM compared with different algorithms, including the average efficiency improvement results within 45 minutes, the minimum efficiency improvement results and the maximum efficiency improvement results. It can be seen that the performance of the dynamically tuned thread pool is significantly improved than that of the static pool, and IFSO-SVM will have an average improvement of 9.12% to 38.00% compared with other comparison algorithms. Therefore, it can be concluded that the intelligent optimization of ICT server performance based on the IFSO-SVM dynamic thread pool algorithm has a good effect.

In other embodiments provided by the present invention, a system for optimizing the performance of the thread pool of a power grid information communication server is provided. As shown in FIG. 8 , the system includes: a functional relationship establishing module 11 for optimizing the performance of the thread pool, a thread pool Tuning the model training module 12 and the thread pool size tuning module 13;

The function relationship establishing module 11 for the optimal thread pool performance is used to analyze the factors affecting the thread pool performance, so as to optimize the thread pool performance and achieve the purpose of optimizing the server performance;

The thread pool tuning model training module 12 is used for inputting the performance test data of the communication server into the thread pool tuning model based on the support vector machine to obtain the hyperparameters of the trained thread pool tuning model;

The thread pool size tuning module 13 is used to judge whether the current thread pool size is the optimal size through the trained support vector machine prediction model, if not, reset the thread pool, and select the thread pool feature data that meets certain conditions to dynamically update training sample set;

The thread pool tuning model is established according to the thread pool performance data throughput, task operation time, task blocking time and the corresponding optimal thread pool size, and the thread pool performance optimization is to select an appropriate thread pool size according to the number of user requests.

The original training sample set is constructed from a large number of ICT server performance experimental data, and the size of the thread pool in different power grid scenarios is dynamically adjusted by the trained support vector machine, and the training sample set is dynamically updated with real-time performance experimental data. , which can realize dynamic and intelligent tuning of the ICT server.

Preferably, in any of the above embodiments, the function relationship establishing module 11 with the best thread pool performance is used to analyze factors affecting the performance of the thread pool, and the steps specifically include:

(1) Let the user task response time be t _response , the queue waiting time of the task in the queue is t _queue , and the task processing time in the pool in the thread pool is t _pool , then t _response = t _queue + t _pool ;

(2) The processing time of a task in the thread pool includes the task preempting CPU operation time t _operation and the waiting time t _{wait for} the task to be suspended due to waiting for system resources, that is, t _pool = t _operation + t _wait . Therefore, the end user task response time t _response = t _queue + t _operation + t _wait ;

(3) Suppose the system throughput is m, the thread pool size is n, and the task operation time is t _operation , then the mathematical model of task queuing time is t _queuing = f(n, m, t _pool ) = f(n, m, t _operation + t _wait );

(4) Assuming that the time consumed by waiting for system resources is blocked as T _blocking , and the time that threads in the pool occupy for CPU operations is T _operation , the mathematical model of task waiting time can be written as t _waiting = g(n, T _operation , T _blocking );

(5) The task operation time t _operation refers to the time consumed by the user task to preempt the CPU to execute the task after entering the thread pool. For each user task, its operation time can be considered as a constant, independent of other parameters such as throughput and thread pool size, t _operation = T _operation ;

(6) To sum up, the mathematical model of user response time reflecting thread pool performance can be built as

_tresponse = _{tqueue+toperation + twait}

=f(n,m,T _operation +g(n,T _operation ,T _blocking ))

+T _operation +g(n, T _operation , T _blocking ),

It can be written as t _response = h(n, m, T _operation , T _blocking );

(7) To optimize the performance of the thread pool, that is, to minimize the user task response time t _response . If the above formula is continuously differentiable, the necessary condition for obtaining the minimum value is _t'response =h'(n _best , m, T _operation , T _blocking )=0.

Preferably, in any of the above embodiments, the support vector machine parameter selection module 12 includes: a support vector machine parameter initialization module and a support vector machine parameter training module;

The support vector machine parameter initialization module is used to initialize the hyperparameters of the support vector machine; wherein, the hyperparameters include: the penalty factor C, the parameter γ of the radial basis kernel function;

The support vector machine training module is used for iterative optimization according to the obtained classification accuracy as the fitness function of IFSO, and finally the optimal hyperparameters are obtained.

By adopting the thread pool tuning model based on IFSO-SVM, a higher classification accuracy is obtained, which improves the optimization effect of FSO, makes it easier to jump out of the local optimum, and intelligently reduces the user response time of the server, especially in the peak of access. It can play the role of peak shaving, which can improve the execution efficiency of the server.

Preferably, in any of the above embodiments, the support vector machine parameter training module is specifically used to calculate the optimal hyperparameters suitable for the thread pool size tuning module 13, and the steps specifically include:

(1) Initialize the position and velocity of each fluid particle, the density of the fluid, the direction of motion, and the normal pressure;

(2) Calculate the objective function value, update the optimal objective function value, the optimal position and the worst objective function value, and calculate the fluid particle density;

(3) Normalize the objective function value and calculate the pressure of the fluid particles;

(4) Calculate the pressure and velocity direction of other fluid particles to the current particle;

(5) Calculate fluid velocity value and velocity vector according to Bernoulli equation;

(6) Update the position of the particle;

(7) Repeat steps (2)-(6) until the termination condition is satisfied.

Among them, in order to improve the accuracy of the fluid search algorithm, a two-stage optimization mechanism is adopted, namely the diversified search in the first stage and the refined exploration in the second stage.

The beneficial effect of adopting the above-mentioned further scheme is: by simplifying the calculation method of the pressure direction, and dividing the optimization process into two stages of diversification search and refined exploration, it avoids that the original FSO is often prone to fall into local optimum, and the amount of calculation is large. It can greatly improve the convergence speed of model training.

Preferably, in any of the above embodiments, the thread pool size tuning module 13 is specifically used to:

Input the real-time running thread pool performance monitoring data as a test sample into the support vector machine to obtain the best thread pool size category to which it belongs;

Determine whether the obtained optimal thread pool size is consistent with the current size, if not, reset the thread pool and dynamically adjust the thread pool size

Judge whether the feature data satisfies the KKT (Karush-Kuhn-Tucker) condition, and if so, replace the point that most violates the KKT condition in the training sample set, and pass it to the support vector machine training module to generate a new classification of the optimal thread pool size. flat.

In the thread pool tuning model based on support vector machine, the KKT condition is used as the judgment condition for the update of the training sample set. For the training sample set, a fixed size is maintained to avoid the infinite sample set with the continuous introduction of new samples. If it increases, the thread pool tuning model can adapt to complex and changeable environments.

The above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be described in the foregoing embodiments. Modifications are made to the technical solutions of the present invention, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A method for optimizing the performance of a thread pool of a power grid information communication server is characterized by comprising the following steps:

s1, analyzing factors influencing the performance of the thread pool, and establishing a thread pool performance model;

s2, inputting the performance test data of the communication server into a thread pool tuning model based on a support vector machine to obtain the hyper-parameters of the trained thread pool tuning model;

s3, judging whether the current thread pool size is the optimal size through the trained support vector machine prediction model, resetting the thread pool if the current thread pool size is not the optimal size, and dynamically updating the training sample set by selecting the thread pool characteristic data meeting certain conditions;

and the thread pool tuning model is established according to the thread pool performance data throughput, the task operation time, the task blocking time and the corresponding optimal thread pool size.

2. The method for optimizing the performance of the thread pool of the grid information communication server according to claim 1, wherein the S1 specifically includes:

(1) let the user task response time be t_{Response to}The queue waiting time of the task in the queue is t_{Queuing up}The processing time of the task in the thread pool is t_PoolThen t is_{Response to}＝t_{Queuing up}+t_Pool；

(2) The processing time of a task in the thread pool comprises the CPU operation time t occupied by the task_OperationsAnd a waiting time t for the task to be suspended by waiting for system resources_{Wait for}I.e. t_Pool＝t_Operations+t_{Wait for}And thus end user task response time t_{Response to}＝t_{Queuing up}+t_Operations+t_{Wait for}；

(3) Setting the system throughput as m, the thread pool size as n, and the task operation time as t_OperationsThen the mathematical model of the task queuing time is t_{Queuing up}＝f(n,m,t_Pool)＝f(n,m,t_Operations+t_{Wait for})；

(4) Let T be the time consumed by waiting for system resources to be blocked_{Blocking of a vessel}The time of CPU operation occupied by the thread in the pool is T_OperationsThen the mathematical model of task latency can be written as t_{Wait for}＝g(n,T_Operations,T_{Blocking of a vessel})；

(5) Task operation time t_OperationsThe method is characterized in that the time consumed by a CPU to execute tasks is preempted after user tasks enter a thread pool, for each user task, the operation time can be regarded as a constant and is irrelevant to other parameters such as throughput, thread pool size and the like, and t_Operations＝T_Operations；

(6) In summary, a mathematical model of user response time reflecting thread pool performance can be constructed as

t_{Response to}＝t_{Queuing up}+t_Operations+t_{Wait for}

＝f(n,m,T_Operations+g(n,T_Operations,T_{Blocking of a vessel}))+T_Operations+g(n,T_Operations,T_{Blocking of a vessel})，

Can be written as t_{Response to}＝h(n,m,T_Operations,T_{Blocking of a vessel})；

(7) To optimize the performance of the thread pool, i.e. to make the user task response time t_{Response to}Taking a minimum value, if the above formula is continuously differentiable, the necessary condition for taking the minimum value is t'_{Response to}＝h'(n_best,m,T_Operations,T_{Blocking of a vessel})＝0。

3. The method for optimizing the performance of the thread pool of the grid information communication server according to claim 1, wherein the S2 specifically includes:

s21, initializing the hyper-parameters of the support vector machine; the hyper-parameters include: penalty factor C, parameter gamma of radial basis kernel function;

and S22, performing cross training by using a support vector machine, and performing iterative optimization by using the obtained classification accuracy as a fitness function of the improved fluid search optimization algorithm to finally obtain the optimal hyperparameter.

4. The method for optimizing the performance of the thread pool of the grid information communication server according to claim 3, wherein the S22 specifically includes:

(1) initializing the position, the speed, the density and the moving direction of each fluid particle and normal pressure;

(2) calculating an objective function value, updating an optimal objective function value, an optimal position and a worst objective function value, and calculating the density of fluid particles;

(3) normalizing the objective function value and calculating the pressure of the fluid particles;

(4) calculating the pressure and the speed direction of other fluid particles to the current particle;

(5) calculating a fluid velocity value and a velocity vector according to a Bernoulli equation;

(6) updating the position of the particle;

(7) and (5) repeating the steps (2) to (6) until a termination condition is met.

5. The method for optimizing the performance of the thread pool of the grid information communication server according to any one of claims 1 to 4, wherein the step S3 specifically includes:

inputting the performance monitoring data of the thread pool running in real time into a support vector machine as a test sample to obtain the size category of the optimal thread pool;

judging whether the size of the obtained optimal thread pool is consistent with the current size, if not, resetting the thread pool, and dynamically adjusting the size of the thread pool;

and judging whether the characteristic data meets the KKT condition or not, if so, replacing the point most violating the KKT condition in the training sample set, and performing training and learning through a support vector machine to obtain new classification hyperplanes of the sizes of all the optimal thread pools.

6. A system for optimizing the performance of a thread pool of a power grid information communication server is characterized by comprising the following steps: the function relation establishing module for the optimal performance of the thread pool, the support vector machine parameter selecting module and the thread pool size optimizing module;

the function relation establishing module with the optimal thread pool performance is used for analyzing factors influencing the thread pool performance so as to optimize the thread pool performance and achieve the aim of optimizing the server performance;

the support vector machine parameter selection module is used for inputting the performance test data of the communication server into a thread pool tuning model based on the support vector machine to obtain the hyperparameter of the trained thread pool tuning model;

the thread pool size optimizing module is used for judging whether the current thread pool size is the optimal size or not through a trained support vector machine prediction model, resetting the thread pool if the current thread pool size is not the optimal size, and dynamically updating a training sample set by using thread pool characteristic data meeting certain conditions;

and the thread pool tuning model is established according to the thread pool performance data throughput, the task operation time, the task blocking time and the corresponding optimal thread pool size.

7. The system for optimizing the performance of the thread pool of the power grid information communication server according to claim 6, wherein the function relationship establishing module with the optimal performance of the thread pool is used for analyzing factors influencing the performance of the thread pool, and the steps specifically include:

(1) let the user task response time be t_{Response to}The queue waiting time of the task in the queue is t_{Queuing up}The processing time of the task in the thread pool is t_PoolThen t is_{Response to}＝t_{Queuing up}+t_Pool；

(2) The processing time of a task in the thread pool comprises the CPU operation time t occupied by the task_OperationsAnd a waiting time t for the task to be suspended by waiting for system resources_{Wait for}I.e. t_Pool＝t_Operations+t_{Wait for}And thus end user task response time t_{Response to}＝t_{Queuing up}+t_Operations+t_{Wait for}；

(3) Setting the system throughput as m, the thread pool size as n, and the task operation time as t_OperationsThen the mathematical model of the task queuing time is t_{Queuing up}＝f(n,m,t_Pool)＝f(n,m,t_Operations+t_{Wait for})；

(4) Let T be the time consumed by waiting for system resources to be blocked_{Blocking of a vessel}The time of CPU operation occupied by the thread in the pool is T_OperationsThen the mathematical model of task latency can be written as t_{Wait for}＝g(n,T_Operations,T_{Blocking of a vessel})；

(5) Task operation time t_OperationsThe method refers to that a user task occupies the time consumed by a CPU to execute the task after entering a thread pool. For each user task, the operation time can be regarded as a constant, and is independent of other parameters such as throughput, thread pool size and the like, and t_Operations＝T_Operations；

(6) In summary, a mathematical model of user response time reflecting thread pool performance can be constructed as

t_{Response to}＝t_{Queuing up}+t_Operations+t_{Wait for}

＝f(n,m,T_Operations+g(n,T_Operations,T_{Blocking of a vessel}))+T_Operations+g(n,T_Operations,T_{Blocking of a vessel})，

Can be written as t_{Response to}＝h(n,m,T_Operations,T_{Blocking of a vessel})；

(7) To optimize the performance of the thread pool, the user task response time t is set_{Response to}Taking a minimum value, if the above formula is continuously differentiable, the necessary condition for taking the minimum value is t'_{Response to}＝h'(n_best,m,T_Operations,T_{Blocking of a vessel})＝0。

8. The system for optimizing the performance of the thread pool of the grid information communication server according to claim 6, wherein the support vector machine parameter selection module comprises: the device comprises a support vector machine parameter initialization module and a support vector machine parameter training module;

the support vector machine parameter initialization module is used for initializing the hyper-parameters of the support vector machine; wherein the hyper-parameters comprise: penalty factor C, parameter gamma of radial basis kernel function;

and the support vector machine training module is used for performing iterative optimization as a fitness function of the improved fluid search optimization algorithm according to the obtained classification accuracy, and finally obtaining the optimal hyperparameter.

9. The system according to claim 8, wherein the support vector machine parameter training module is specifically configured to calculate an optimal hyper-parameter applicable to the thread pool size tuning module, and the method specifically includes the following steps:

(1) initializing the position, the speed, the density and the moving direction of each fluid particle and normal pressure;

(2) calculating an objective function value, updating an optimal objective function value, an optimal position and a worst objective function value, and calculating the density of fluid particles;

(3) normalizing the objective function value and calculating the pressure of the fluid particles;

(4) calculating the pressure and the speed direction of other fluid particles to the current particle;

(5) calculating a fluid velocity value and a velocity vector according to a Bernoulli equation;

(6) updating the position of the particle;

(7) and (5) repeating the steps (2) to (6) until a termination condition is met.

10. The system for optimizing the performance of the thread pool of the grid information communication server according to any one of claims 6 to 9, wherein the thread pool size optimizing module is specifically configured to:

inputting the performance monitoring data of the thread pool running in real time into a support vector machine as a test sample to obtain the size category of the optimal thread pool;

judging whether the size of the obtained optimal thread pool is consistent with the current size, if not, resetting the thread pool, and dynamically adjusting the size of the thread pool;

and judging whether the characteristic data meets the KKT condition or not, if so, substituting the point most violating the KKT condition in the training sample set, and handing the point to a support vector machine training module to generate a new classification hyperplane of each optimal thread pool size.

Images