CN109992412B - Capacity adjusting method and device of cloud server, storage medium and cloud server - Google Patents
Capacity adjusting method and device of cloud server, storage medium and cloud server Download PDFInfo
- Publication number
- CN109992412B CN109992412B CN201910122494.3A CN201910122494A CN109992412B CN 109992412 B CN109992412 B CN 109992412B CN 201910122494 A CN201910122494 A CN 201910122494A CN 109992412 B CN109992412 B CN 109992412B
- Authority
- CN
- China
- Prior art keywords
- capacity
- cloud server
- predicted
- neural network
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention relates to a capacity adjusting method and device of a cloud server, a storage medium and the cloud server. The capacity adjusting method of the cloud server comprises the following steps: acquiring time sequence data of the capacity load of the cloud server before a time node to be predicted; inputting the time sequence data into a first neural network model to obtain a first prediction capacity of the cloud server at the time node to be predicted; inputting the time sequence data into a second neural network model, acquiring a second prediction capacity of the cloud server at the time node to be predicted, and acquiring the prediction capacity of the cloud server at the time node to be predicted. According to the capacity adjusting method of the cloud server, the capacity required by the cloud server at the time node to be predicted is predicted according to the time sequence data of the capacity load of the cloud server before the time node to be predicted, so that the capacity of the cloud server can be adjusted in advance, and the normal work of the application service of the cloud server is guaranteed.
Description
Technical Field
The invention relates to the technical field of cloud servers, in particular to a capacity adjusting method and device of a cloud server, a storage medium and the cloud server.
Background
The cloud server is a network-based high-availability computing mode, a physical server cluster which is across servers and even across data centers is virtualized, a host product which can rapidly dispatch high-availability computing resources is supported, the configuration of the service scale and the service capacity of the host can be configured according to the needs of users, and the configuration can be flexibly adjusted.
The capacity of a traditional cloud server is distributed in a static distribution mode, and the capacity is distributed according to the maximum load of an application service. However, when the load of the cloud server is in a peak period, the related application service cannot be normally performed due to insufficient capacity resource supply; when the load of the cloud server is in the valley period, the capacity resource is wasted because the supply and demand of the capacity resource are greater than the demand.
The existing elastic cloud server can allocate required capacity resources for the application service in real time according to the condition of the application service load of the existing elastic cloud server, and adjust the capacity of the cloud server. But the elastic cloud server has the problem that the resource allocation is time-consuming. The elastic cloud server adjusts the capacity of the cloud server when the current relevant application service load is large and more resources need to be allocated through real-time analysis, and a certain time is needed from the cloud capacity resource allocation to the cloud capacity resource usage, so that time delay exists. This still causes the relevant application service to fail to operate normally in a short time, resulting in monetary losses for the user and the cloud server provider.
Disclosure of Invention
Based on this, the invention aims to provide a capacity adjusting method for a cloud server, which can predict the capacity of the cloud server required by a time node to be predicted according to time sequence data of capacity load of the cloud server before the time node to be predicted, so that the capacity of the cloud server can be adjusted in advance, and normal work of application service of the cloud server is ensured.
The capacity adjusting method of the cloud server is implemented by the following scheme:
a capacity adjusting method of a cloud server comprises the following steps:
acquiring time sequence data of the capacity load of the cloud server before a time node to be predicted;
inputting the time sequence data into a first neural network model to obtain a first predicted capacity of the cloud server at the time node to be predicted, wherein the first neural network model is trained by the time sequence data of the capacity load of the cloud server in a first time interval before the time node to be predicted; the training process of the first neural network model comprises the following steps: acquiring a time series data set of the capacity load of the cloud server in a first time interval; training a first neural network model by using a time series dataset of the cloud server capacity load in the first time interval, and optimizing weight and threshold values of the first neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm;
inputting the time sequence data into a second neural network model, and acquiring second predicted capacity of the cloud server at the time node to be predicted, wherein the second neural network model is trained by the time sequence data of the capacity load of the cloud server in a second time interval before the time node to be predicted;
and acquiring the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity.
According to the capacity adjusting method of the cloud server, two different prediction capacities are obtained through two neural network models trained in different time intervals according to time sequence data of the capacity load of the cloud server before the time node to be predicted, the capacity required by the cloud server at the time node to be predicted is obtained according to the two prediction capacities, the prediction capacity of the cloud server can be obtained more accurately, and therefore the capacity of the cloud server can be adjusted in advance, and normal work of application service of the cloud server is guaranteed.
In one embodiment, the start time of the first time interval is earlier than the second time interval. And the output second prediction capacity of the first neural network model can reflect the current trend of the cloud server capacity according to the latest trend of the cloud server capacity.
In order to calculate the predicted capacity more accurately, in an embodiment, obtaining the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity includes:
acquiring a first weight value of a first prediction capacity;
acquiring a second weight value of a second prediction capacity;
and according to the first prediction capacity and the first weight value, and the second prediction capacity and the second weight value, performing weighted calculation on the first prediction capacity and the second prediction capacity to obtain the prediction capacity of the cloud server at the time node to be predicted.
In order to more accurately perform weighting calculation on the first predicted capacity, in an embodiment, the obtaining a first weight value of the first predicted capacity specifically includes:
and acquiring the first weight value according to one or more combinations of the number of the application services accessed by the cloud server, the number of users of the application services and the activity time of the application services in a first time period.
In order to more accurately perform weighting calculation on the second predicted capacity, in an embodiment, the obtaining a second weight value of the second predicted capacity specifically includes:
and acquiring the second weight value according to the combination of one or more of the number of the application services accessed by the cloud server, the number of the application service users and the application service activity time in a second time period.
In one embodiment, after obtaining the predicted capacity of the time node to be predicted, the method further includes the following steps:
and adjusting the capacity of the cloud server at the time node to be predicted according to the predicted capacity, so that the capacity of the cloud server is adjusted in advance, and the normal work of the application service of the cloud server at the time node to be predicted is ensured.
If the difference between the actual demand capacity and the predicted capacity is too large, the application service may not operate normally or the capacity resource may be wasted, and therefore, in one embodiment, the method further includes the following steps:
acquiring the actual demand capacity of the cloud server at the time node to be predicted;
obtaining a difference between the actual demand capacity and the predicted capacity;
and if the difference between the actual required capacity and the predicted capacity is larger than a first set threshold, adjusting the capacity of the cloud server according to the difference.
In one embodiment, it is also possible to notify the administrator to perform capacity adjustment, including the following steps:
acquiring the actual demand capacity of the cloud server at the time node to be predicted;
obtaining a difference between the actual demand capacity and the predicted capacity;
if the difference between the actual demand capacity and the predicted capacity is greater than a first set threshold, the difference is sent to an administrator client.
Specifically, in one embodiment, training a first neural network model using the time series data set of the cloud server capacity load in the first time interval, and optimizing the weight and the threshold of the first neural network model through a horizontal-vertical double-chord optimization algorithm includes:
initializing a first neural network model, and setting an initial weight and a threshold;
initializing an artificial particle population, and carrying out individual coding on the artificial particle population;
training a first neural network model by using the capacity load time series data of the cloud server in a first time interval, and evaluating the fitness value of each individual of the artificial particle population by using a training sample error;
and performing horizontal double-chord global search and vertical double-chord local development on the artificial particle population, and preferentially reserving the generated new individuals and parent individuals.
In one embodiment, when training the first neural network model, the method further includes the step of judging the training end:
and if the error of the training sample is smaller than a second set threshold value or the iteration times of the horizontal and vertical double-chord optimizing algorithm exceed the maximum set iteration times, stopping training the first neural network model, and outputting the current weight and threshold value as the weight and threshold value of the first neural network model.
In one embodiment, the training process of the second neural network model comprises:
acquiring a time sequence data set of the cloud server capacity load in a second time interval;
training a second neural network model by using the time series data set of the cloud server capacity load in the second time interval, and optimizing the weight and the threshold of the second neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm.
Specifically, in one embodiment, training a second neural network model using the time-series data set of the cloud server capacity load in the second time interval, and optimizing the weight and the threshold of the second neural network model through a horizontal-vertical double-chord optimization algorithm includes:
initializing a second neural network model, and setting an initial weight and a threshold;
initializing an artificial particle population, and carrying out individual coding on the artificial particle population;
training a second neural network model by using the capacity load time series data of the cloud server in a second time interval, and evaluating the fitness value of each individual of the artificial particle population by using the training sample error;
and performing horizontal double-chord global search and vertical double-chord local development on the artificial particle population, and preferentially reserving the generated new individuals and parent individuals.
In one embodiment, when training the second neural network model, the method further comprises the step of judging the training end:
and if the error of the training sample is smaller than a third set threshold value or the iteration times of the horizontal and vertical double-chord optimizing algorithm exceed the maximum set iteration times, stopping training the second neural network model, and outputting the current weight and the threshold value as the weight and the threshold value of the second neural network model.
Further, the present invention also provides a capacity adjustment device for a cloud server, including:
the data acquisition module is used for acquiring time sequence data of the capacity load of the cloud server before the time node to be predicted;
the first input module is used for inputting the time series data into a first neural network model and acquiring first predicted capacity of the cloud server at the time node to be predicted, wherein the first neural network model is trained by the time series data of the capacity load of the cloud server in a first time interval before the time node to be predicted; the training process of the first neural network model comprises the following steps: acquiring a time series data set of the capacity load of the cloud server in a first time interval; training a first neural network model by using a time series dataset of the cloud server capacity load in the first time interval, and optimizing weight and threshold values of the first neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm;
the second input module is used for inputting the time series data into a second neural network model and acquiring second predicted capacity of the cloud server at the time node to be predicted, wherein the second neural network model is trained by the time series data of the capacity load of the cloud server in a second time interval before the time node to be predicted;
and the prediction module is used for acquiring the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity.
According to the capacity adjusting device of the cloud server, two different prediction capacities are obtained through two neural network models trained in different time intervals according to the time sequence data of the capacity load of the cloud server before the time node to be predicted, the capacity required by the cloud server at the time node to be predicted is obtained according to the two prediction capacities, the prediction capacity of the cloud server can be obtained more accurately, and therefore the capacity of the cloud server can be adjusted in advance, and normal work of application service of the cloud server is guaranteed.
Further, the present invention also provides a computer readable medium, on which a computer program is stored, which when executed by a processor implements the capacity adjustment method of the cloud server as described in any one of the above embodiments.
Further, the present invention also provides a cloud server, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein:
when the processor executes the computer program, the capacity adjustment method of the cloud server according to any one of the above embodiments is implemented.
For a better understanding and practice, the invention is described in detail below with reference to the accompanying drawings.
Drawings
FIG. 1 is a schematic block diagram of an application environment configuration of a capacity adjustment method of a cloud server in one embodiment;
FIG. 2 is a flow diagram of a method for capacity adjustment of a cloud server in one embodiment;
fig. 3 is a flowchart of obtaining the predicted capacity of the cloud server at the time node to be predicted in one embodiment;
FIG. 4 is a flow diagram of a cloud server adjusting capacity in real time in one embodiment;
FIG. 5 is a flow diagram of a training of a first neural network model in one embodiment;
FIG. 6 is a diagram illustrating a first neural network model according to one embodiment;
FIG. 7 is a schematic diagram of a transverse two-chord optimization algorithm in one embodiment;
FIG. 8 is a schematic diagram of an embodiment of a horizontal-vertical bichord optimization algorithm;
FIG. 9 is a detailed flow diagram of a first neural network model training in one embodiment;
FIG. 10 is a schematic diagram of the structures of a first neural network model and a second neural network model in one embodiment;
FIG. 11 is a diagram illustrating an exemplary configuration of a capacity adjustment mechanism of a cloud server;
fig. 12 is a schematic structural diagram of a cloud server in one embodiment.
Detailed Description
Referring to fig. 1, fig. 1 is a schematic block diagram of an application environment structure of a capacity adjustment method of a cloud server according to an embodiment of the present invention, where the application environment of the capacity adjustment method of the cloud server according to the embodiment is a cloud server 20. The cloud server 20 may run one or more application services, and provide capacity resources required for normal operation for the one or more application services from the capacity resources of the cloud server. As shown in fig. 1, a client 10 connects to a cloud server 20 through the internet and creates or accesses one or more application services on the cloud server 20. The client 10 may be a desktop computer 101, a portable computer 102, a smart phone 103, other smart devices, and the like.
In other embodiments, the cloud server 20 may be not only a single server, but also a server cluster composed of multiple servers.
Referring to fig. 2, in an embodiment, a capacity adjustment method of a cloud server includes the following steps:
step S201: and acquiring time sequence data of the capacity load of the cloud server before the time node to be predicted.
In this embodiment, the measurement units of the time node to be predicted and the time series data are all one hour, and in other embodiments, the measurement units may be several hours or days.
The time sequence data of the cloud server capacity load before the time node to be predicted is a set of the cloud server capacity load values which are arranged in time sequence and are arranged in time sequence several hours before the time node to be predicted.
The capacity load of the cloud server may be an average value of the capacity loads of the cloud servers within one hour, or may be a maximum value of the capacity loads of the cloud servers within one hour, and in this embodiment, the maximum value of the capacity loads of the cloud servers within one hour is preferred.
Step S202: and inputting the time sequence data into a first neural network model, and acquiring first predicted capacity of the cloud server at the time node to be predicted, wherein the first neural network model is trained by the time sequence data of the capacity load of the cloud server in a first time interval before the time node to be predicted.
Step S203: inputting the time sequence data into a second neural network model, and acquiring second predicted capacity of the cloud server at the time node to be predicted, wherein the second neural network model is trained by the time sequence data of the capacity load of the cloud server in a second time interval before the time node to be predicted.
In this embodiment, the first neural network model and the second neural network model adopt a bp (back propagation) neural network, preferably an Elman neural network, and in other embodiments, other neural networks may also be used.
Step S204: and acquiring the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity.
The predicted capacity is obtained by calculating the first predicted capacity and the second predicted capacity, and the current trend is predicted by comprehensively combining the historical trend and the recent trend of the capacity load of the cloud server.
The predicted capacity may be an average value of the capacity load of the cloud server within one hour of the time node to be predicted, or may be a maximum value of the capacity load of the cloud server within one hour of the time node to be predicted, and in this embodiment, it is preferable that the maximum value of the capacity load of the cloud server within one hour of the time node to be predicted.
According to the capacity adjusting method of the cloud server, two different prediction capacities are obtained through two neural network models trained in different time intervals according to time sequence data of the capacity load of the cloud server before the time node to be predicted, the capacity required by the cloud server at the time node to be predicted is obtained according to the two prediction capacities, the prediction capacity of the cloud server can be obtained more accurately, and therefore the capacity of the cloud server can be adjusted in advance, and normal work of application service of the cloud server is guaranteed.
In one embodiment, the first time interval is a period of time, preferably a year, before the time node to be predicted, that is, time series data of a capacity load of a cloud server in a period of one year before the time node to be predicted is selected to train the first neural network. In other embodiments, the first time interval may also be a time interval range longer or shorter than one year, and the second time interval may also be a time interval range longer or shorter than 30 days, so that the starting time of the first time interval is earlier than the second time interval, that is, the first time interval is longer than the second time interval. The ending time of the first time interval and the ending time of the second time interval are preferably the same, and may be the time closest to the time node to be predicted, or may be a designated time before the time node to be predicted.
The first neural network model is trained through time series data of the cloud server capacity load in the last year, the output first prediction capacity of the first neural network model can reflect the current trend of the cloud server capacity according to the historical trend of the cloud server capacity, the second neural network model is trained through time series data of the cloud server capacity load in the last 30 days, and the output second prediction capacity of the second neural network model can better reflect the current trend of the cloud server capacity according to the latest trend of the cloud server capacity.
In order to calculate the predicted capacity more accurately, in an embodiment, obtaining the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity includes:
step S301: a first weight value of the first prediction capacity is obtained.
Step S302: and acquiring a second weight value of the second prediction capacity.
The first weight value and the second weight value are obtained through one or more combinations of the number of application services, the number of application service users and the application service activity time accessed by the cloud server in the last year and last 30 days respectively. The application service number is an average value of the application service numbers in the cloud server in the last year and the last 30 days, the application service user number is an average value of the user numbers of all the application services in the cloud server in the last year and the last 30 days, the activity time of the application service includes the activity time of the game, and the user number changes obviously in the activity time range, so that the activity time of the application service is included in the calculation range of the weight value.
Preferably, the first weight value includes parameters such as the number of application services, the number of users of the application services, and the application service activity time accessed by the cloud server in the last year, and the second weight value includes parameters such as the number of application services, the number of users of the application services, and the application service activity time accessed by the cloud server in the last 30 days.
Step S303: and according to the first prediction capacity and the first weight value, and the second prediction capacity and the second weight value, performing weighted calculation on the first prediction capacity and the second prediction capacity to obtain the prediction capacity of the cloud server at the time node to be predicted.
Suppose the first weight values are respectively delta1，δ2…, δ n, the second weight value is ω1,ω2,…，ωm. Setting the prediction capacity of a time node to be predicted as C, and setting the first prediction capacity as CyearThe second predicted capacity is CdayThen, the calculation formula of the predicted capacity C is:
in one embodiment, after the predicted capacity C is calculated, the capacity of the cloud server is adjusted in advance according to the adjustment of the capacity of the cloud server at the time node to be predicted, so that the application service of the cloud server at the time node to be predicted works normally.
Referring to fig. 4, in an embodiment, when the cloud server adjusts the capacity of the cloud server in advance and runs at the time node to be predicted with the predicted capacity, the following steps are further performed:
step S401: and acquiring the actual demand capacity of the cloud server at the time node to be predicted.
Step S402: a difference between the actual demand capacity and the predicted capacity is obtained.
Step S403: and if the difference between the actual required capacity and the predicted capacity is larger than a first set threshold, adjusting the capacity of the cloud server according to the difference.
The actual demand capacity of the cloud server at the time node to be predicted is obtained by the cloud server according to capacity resource calculation required by the currently running application service, and if the difference between the actual demand capacity and the predicted capacity is too large, it indicates that the cloud server is currently in a state of too large or too small capacity, which causes resource waste or abnormal operation of related application services. The first set threshold is set in advance according to the performance of the cloud server.
In other embodiments, the difference value may be sent to an administrator client, which may be an administrator account or a mailbox bound to the administrator account, and notifies an administrator to perform real-time adjustment on the capacity of the cloud server.
Referring to fig. 5, in one embodiment, the training process of the first neural network model includes the following steps:
step S501: and acquiring a time series data set of the cloud server capacity load in a first time interval.
Step S502: training a first neural network model by using the time series data set of the cloud server capacity load in the first time interval, and optimizing the weight and the threshold of the first neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm.
The first time interval is one year, and in other embodiments, the first time interval may also be other longer or shorter time intervals. Collecting the time series data S of the cloud server capacity load in one yearyearInputting the predicted cloud service container demand capacity C corresponding to the time node t into the constructed first neural network modelyearAccording to the required capacity CyearAnd training sample error MSE of actual capacity, and optimizing the weight W and the threshold value theta of the first neural network model by using a horizontal-vertical double-chord optimizing algorithm.
Specifically, referring to fig. 6, fig. 6 is a schematic structural diagram of a first neural network model, which is an Elman neural network, and is a BP network with feedback, which has a local memory unit and a forward neural network connected with local feedback. Compared with the traditional BP neural network, the Elman neural network has stronger dynamic behavior and computing power, is suitable for establishing a prediction model of a time sequence, and is beneficial to predicting the capacity load of the cloud server in a short time.
The key parameters of the first neural network model are set as follows: the hierarchical structure is four layers, namely an input layer, a hidden layer, a carrying layer and an output layer. The number of nodes of the input layer is n, the number of nodes of the hidden layer is h, the number of nodes of the receiving layer is h, and the number of nodes of the output layer is m. IWnhWeights, CW, for the n-th node of the input layer to be connected to the h-th node of the hidden layerhhFor the weight, OW, of the h-th node of the hidden layer connected to the h-th node of the bearer layerhmFor the h node of the hidden layer to be connected to the m node of the output layer, bh(i＝1,...,h)The threshold value of the ith node of the hidden layer. Wherein, the threshold value theta and the weight value W are initial values of 0,1]Random numbers over the interval.
The traditional Elman neural network adopts a momentum gradient descent method to adjust the weight W and the threshold theta, so that the Elman neural network is easy to fall into a local optimal solution in the prediction process, the accurate value of a prediction result is reduced, and when influence factors and learning samples increase, the calculated amount and the weight number of the Elman neural network increase rapidly, so that the convergence speed is slow. Therefore, in this embodiment, a longitudinal and transverse double-chord Optimization Algorithm (CCSOA) is added in the Elman training to optimize the first neural network model, and the threshold and weight of the neural network are adaptively adjusted through longitudinal and transverse two-dimensional degree double-chord Optimization, so as to improve the prediction accuracy and speed.
The transverse and longitudinal double-chord optimization algorithm is a novel meta-heuristic group intelligent algorithm which is provided according to the three-dimensional motion of natural particles based on sine functions and cosine functions, can simulate the three-dimensional motion of the natural particles based on the sine functions and the cosine functions under the constraint condition omega, carries out iterative search calculation for many times, solves the optimal solution of an objective function f (x) in a search space S, and rapidly and accurately obtains the optimal value in a limited range through iterative calculation for limited times in a certain limited range.
The horizontal-vertical double-chord optimization algorithm comprises a horizontal double-chord optimization algorithm, please refer to fig. 7, wherein in the horizontal dimension, the population individuals are random individuals P in the horizontal dimension of the populationhrandAnd (4) carrying out curve motion of a sine function for main body guidance to search a global unknown optimal solution. At the same time, the population individuals are selected as the optimal individual P in the transverse dimension of the populationhbestFor auxiliary guidance, the curve motion of a cosine function is performed to reduce the optimization range, guide and correct the optimization direction of random individuals, avoid deviation from the optimal solution, accelerate the optimization speed of the algorithm and ensure the optimization efficiency of the population in the transverse dimension. The calculation formula (1) is as follows:
Ph(t+1)＝Ph(t)+a·sin(r1)·Dhsin+b·cos(r2)·Dhcos (1)
wherein, Ph(t +1) is the iteratively updated position of the individual transverse dimension, Ph(t) is the current position of the individual lateral dimension. r is1、r2Is taken to be [0,2 pi ]]The random number of (2).
Where T is the current number of iterations and T is the total number of iterations. DhsinIs an individual Ph(t) distance from population transverse dimensionRandom individual P in (1)hrandDistance of (D)hcosIs an individual Ph(t) optimal individual P in the transverse dimension from the populationhbestThe distance of (c). DhsinAnd DhcosAs shown in equation (2):
in the formula (2), α and β are random numbers with values in the range of [ -1,1] to enhance the randomness of the population individuals.
In the formula (1) and the formula (2), a is gradually reduced along with the increase of the iteration number t, and the random individual PhrandTo Ph(t +1) is weakened; b becomes gradually larger and the optimal individual PhbestTo PhAnd (t +1) the influence is enhanced, so that the individuals in the population gradually gather to the same global optimal solution, and the convergence of the algorithm is ensured.
As shown in FIG. 7, in the formula (1), the sine function is random individual PhrandIs guided by the main body with a guide coordinate PsinThe search range is large, and the searching optimization blind spot is easy to dig; cosine function to optimize individual PhbestFor auxiliary guidance, the guide coordinate is PcosCan correct the optimizing direction P of the sine random individualsin. Population of individuals P (t) at PsinAnd PcosUnder the complementary action of the two navigation directions, the next iteration position P (t +1) is quickly calculated, so that the algorithm can quickly and accurately find out the optimal individual of the next iteration.
The horizontal-vertical double-chord optimization algorithm also comprises a vertical double-chord optimization algorithm, please refer to fig. 8, and individuals in the population are optimized in the vertical dimension in addition to the double-chord function optimization in the horizontal dimension. The calculation formula (3) is as follows, and the longitudinal dimension is normalized before calculation.
Pv(t+1)＝Pv(t)+a·sin(r1)·Dvsin+b·cos(r2)·Dvcos (3)
Wherein, Pv(t +1) is the position after the iterative update of the longitudinal dimension of the individual,Pv(t) is the current position of the individual longitudinal dimension. r is1、r2Is taken to be [0,2 pi ]]The random number of (2).
Where T is the current number of iterations and T is the total number of iterations. DvsinIs an individual Pv(t) random individuals P in the longitudinal dimension from the populationvrandDistance of (D)vcosIs an individual Pv(t) optimal Individual P in longitudinal dimension from populationvbestThe distance of (c). DvsinAnd DvcosAs shown in equation (4):
in the formula (4), α and β are random numbers having values in the range of [ -1,1 ]. Longitudinal dimension optimization is increased, particle individual information can be widely spread, the transversely stagnated particle individuals can jump out of local optimum through longitudinal calculation, and the algorithm is prevented from being premature.
The traditional group intelligent algorithm only conducts space search optimization of the transverse dimension, so that the calculation of the algorithm is only limited to the transverse dimension of the space. The population information is not in the full communication space of the whole population, and optimization blind spots exist in the iterative computation process of individuals with different transverse dimensions easily. Therefore, on the basis of transverse double-chord optimization, longitudinal dimension optimization is increased, a single-dimension optimization mode of population individuals is changed, greedy competition is introduced, and P after iterative update is performedh(t +1) and Pv(t +1) competes with the parent individuals, and preferentially retains.
From fig. 8, it can be found that the horizontal and vertical two-dimensional double-chord optimization enables individual information to be spread in a chain manner to the whole population, and accuracy and speed of optimization calculation are improved. The navigation guidance of the longitudinal and transverse bidirectional dimension information can self-adaptively correct the optimization calculation direction and position the optimal solution, thereby not only accelerating the optimization speed of the transverse and longitudinal double-chord optimization algorithm, but also avoiding the algorithm from falling into the local optimal solution and improving the solving precision of the algorithm.
Referring to fig. 9, in one embodiment, the step of training the first neural network model includes:
step S901: initializing a first neural network model, and setting an initial weight W and a threshold theta, wherein the weight W and the threshold theta are random numbers of which the initial values are in a [0,1] interval.
Step S902: initializing an artificial particle population, the individual codes of which are P:
Pi＝1,...,N＝[IW11...IWnhCW11...CWhhOW11...OWhmb1...bh]
step S903: training a first neural network model by using capacity load time series data of the cloud server within one year, evaluating the fitness value of each individual of a population according to an equation (1) by using a training sample error MSE, and calculating the fitness value fobj of the individual of the population by using a calculation equation:
wherein p istIs the actual output of the Elman neural network,
and N is the total number of training samples for the target output of the Elman neural network.
Step S904: and (3) performing horizontal double-chord global search according to the formula (1) and performing vertical double-chord local development according to the formula (3), so that new individuals compete with parent individuals, and the preference is reserved.
Step S905: calculating whether the training sample error MSE is smaller than a second set threshold value, for example, 0.01, or whether the iteration number is larger than a set maximum iteration number T, if yes, executing step S906, and if not, executing step S903, wherein the second set threshold value is a set value and can be set according to actual conditions.
Step S906: and (5) ending the training, and outputting the optimal weight W and the threshold value theta to the first neural network model.
Referring to fig. 10, the structure and training process of the second neural network model are substantially the same as those of the first neural network model, and the second neural network model is trained using the time-series data set of the cloud server capacity load in a second time interval, wherein the second time interval is 30 days, and in other embodiments, the first time interval may be other longer or shorter time intervals.
And the second neural network model optimizes the weight W and the threshold value theta through a horizontal-vertical double-chord optimizing algorithm, wherein the hierarchical structure of the second neural network model is the same as that of the first neural network model, but the number of nodes of each hierarchy can be adjusted according to actual needs.
When the second neural network model is trained, whether the training sample error MSE is smaller than a third set threshold value, such as 0.01, or whether the iteration number is larger than a set maximum iteration number T is calculated, wherein the third set threshold value is a set value and can be set according to actual conditions.
Based on the same inventive concept, the present invention further provides a capacity adjustment apparatus of a cloud server, so as to implement the capacity adjustment method of the cloud server, please refer to fig. 11, in an embodiment, the capacity adjustment apparatus 300 of the cloud server of the present invention includes:
the data acquiring module 301 is configured to acquire time series data of a capacity load of the cloud server before a time node to be predicted.
A first input module 302, configured to input the time series data into a first neural network model, and obtain a first predicted capacity of the cloud server at the time node to be predicted, where the first neural network model is trained by the time series data of the cloud server capacity load in a first time interval before the time node to be predicted.
A second input module 303, configured to input the time series data into a second neural network model, and obtain a second predicted capacity of the cloud server at the time node to be predicted, where the second neural network model is trained by the time series data of the cloud server capacity load in a second time interval before the time node to be predicted.
And the predicting module 304 is configured to obtain the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity.
According to the capacity adjusting device of the cloud server, two different prediction capacities are obtained through two neural network models trained in different time intervals according to the time sequence data of the capacity load of the cloud server before the time node to be predicted, the capacity required by the cloud server at the time node to be predicted is obtained according to the two prediction capacities, the prediction capacity of the cloud server can be obtained more accurately, and therefore the capacity of the cloud server can be adjusted in advance, and normal work of application service of the cloud server is guaranteed.
In one embodiment, the start time of the first time interval is earlier than the second time interval.
In one embodiment, the prediction module 304 includes:
a first weight value obtaining unit configured to obtain a first weight value of the first prediction capacity.
A second weight value obtaining unit configured to obtain a second weight value of the second prediction capacity.
And the weighting calculation unit is used for carrying out weighting calculation on the first prediction capacity and the second prediction capacity according to the first prediction capacity and the first weight value as well as the second prediction capacity and the second weight value, and acquiring the prediction capacity of the cloud server at the time node to be predicted.
In one embodiment, the first weight value obtaining unit obtains the first weight value according to a combination of one or more of the number of application services accessed by the cloud server, the number of application service users, and application service active time during a first time period; the second weight value obtaining unit obtains the second weight value according to a combination of one or more of the number of application services, the number of application service users, and the application service active time accessed by the cloud server in a second time period.
In one embodiment, the system further comprises a first adjusting module, configured to adjust the capacity of the cloud server at the time node to be predicted according to the predicted capacity.
In one embodiment, further comprising:
and the actual demand capacity acquisition module is used for acquiring the actual demand capacity of the cloud server at the time node to be predicted.
A difference acquisition module to acquire a difference between the actual demand capacity and the predicted capacity.
And the second adjusting module is used for adjusting the capacity of the cloud server according to the difference value if the difference value between the actual required capacity and the predicted capacity is larger than a first set threshold value.
A sending module, configured to send a difference between the actual required capacity and the predicted capacity to an administrator client if the difference is greater than a first set threshold.
In one embodiment, the method further comprises a first neural network training module for training the first neural network model, comprising:
the first data acquisition unit is used for acquiring a time series data set of the cloud server capacity load in a first time interval.
The first training unit is used for training a first neural network model by using the time sequence dataset of the cloud server capacity load in the first time interval, and optimizing weight values and threshold values of the first neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm.
In one embodiment, the first training unit comprises:
and the first neural network model initialization unit is used for initializing the first neural network model and setting initial weight and threshold values.
And the first artificial particle population initializing unit is used for initializing the artificial particle population and carrying out individual coding on the artificial particle population.
And the first evaluation unit is used for training a first neural network model by using the capacity load time series data of the cloud server in a first time interval and evaluating the fitness value of each individual of the artificial particle population by using the training sample error.
And the first optimization unit is used for performing horizontal double-chord global search and vertical double-chord local development on the artificial particle population and preferentially retaining the generated new individuals and parent individuals.
In one embodiment, the first neural network training module further comprises:
and the first finishing unit is used for stopping training the first neural network model and outputting the current weight and threshold as the weight and threshold of the first neural network model if the error of the training sample is smaller than a second set threshold or the iteration number of the horizontal-vertical double-chord optimizing algorithm exceeds the maximum set iteration number.
In one embodiment, a second neural network training module for training a second neural network model is further included, comprising:
and the second data acquisition unit is used for acquiring a time series data set of the capacity load of the cloud server in a second time interval.
And the second training unit is used for training a second neural network model by using the time series data set of the cloud server capacity load in the second time interval, and optimizing the weight and the threshold of the second neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm.
In one embodiment, the second training unit comprises:
and the second neural network model initialization unit is used for initializing the second neural network model and setting initial weight and threshold values.
And the second artificial particle population initializing unit is used for initializing the artificial particle population and carrying out individual coding on the artificial particle population.
And the second evaluation unit is used for training a second neural network model by using the capacity load time series data of the cloud server in a second time interval, and evaluating the fitness value of each individual of the artificial particle population by using the training sample error.
And the second optimization unit is used for performing horizontal double-chord global search and vertical double-chord local development on the artificial particle population and preferentially retaining the generated new individuals and parent individuals.
In one embodiment, the second neural network training module further comprises:
and the second ending unit is used for stopping training the second neural network model and outputting the current weight and threshold as the weight and threshold of the second neural network model if the error of the training sample is smaller than a third set threshold or the iteration number of the horizontal-vertical double-chord optimization algorithm exceeds the maximum set iteration number.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The present invention also provides a computer-readable medium having stored thereon a computer program which, when executed by a processor, implements the capacity adjustment method of the cloud server in any one of the above embodiments.
Referring to fig. 12, in an embodiment, a cloud server 400 of the present invention includes a memory 401 and a processor 402, and a computer program stored in the memory 401 and executable by the processor 402, where the processor 402 executes the computer program to implement the capacity adjustment method of the cloud server in any one of the above embodiments.
In this embodiment, the processor 402 and the memory 401 are connected by a bus, and the processor 402 may be one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors, or other electronic components. The memory 401 may take the form of a computer program product embodied on one or more storage media including, but not limited to, disk storage, CD-ROM, optical storage, and the like, in which program code is embodied. Computer readable storage media, which include both non-transitory and non-transitory, removable and non-removable media, may implement any method or technology for storage of information. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of the storage medium of the computer include, but are not limited to: phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technologies, compact disc read only memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, magnetic cassettes, magnetic tape storage or other magnetic storage devices, or any other non-transmission medium, which may be used to store information that may be accessed by a computing device, in this embodiment, the processor 402 may also be multiple ones, and the cloud server 400 may also be a server cluster of multiple computers.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.
Claims (16)
1. A capacity adjustment method of a cloud server is characterized by comprising the following steps:
acquiring time sequence data of the capacity load of the cloud server before a time node to be predicted;
inputting the time sequence data into a first neural network model to obtain a first predicted capacity of the cloud server at the time node to be predicted, wherein the first neural network model is trained by the time sequence data of the capacity load of the cloud server in a first time interval before the time node to be predicted; the training process of the first neural network model comprises the following steps: acquiring a time series data set of the capacity load of the cloud server in a first time interval; training a first neural network model by using a time series dataset of the cloud server capacity load in the first time interval, and optimizing weight and threshold values of the first neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm;
inputting the time sequence data into a second neural network model, and acquiring second predicted capacity of the cloud server at the time node to be predicted, wherein the second neural network model is trained by the time sequence data of the capacity load of the cloud server in a second time interval before the time node to be predicted;
and acquiring the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity.
2. The method for adjusting the capacity of the cloud server according to claim 1, wherein:
the start time of the first time interval is earlier than the second time interval.
3. The capacity adjustment method of the cloud server according to claim 1 or 2, wherein obtaining the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity includes:
acquiring a first weight value of a first prediction capacity;
acquiring a second weight value of a second prediction capacity;
and according to the first prediction capacity and the first weight value, and the second prediction capacity and the second weight value, performing weighted calculation on the first prediction capacity and the second prediction capacity to obtain the prediction capacity of the cloud server at the time node to be predicted.
4. The capacity adjustment method of the cloud server according to claim 3, wherein obtaining the first weight value of the first predicted capacity specifically includes:
and acquiring the first weight value according to one or more combinations of the number of the application services accessed by the cloud server, the number of users of the application services and the activity time of the application services in a first time period.
5. The capacity adjustment method of the cloud server according to claim 3, wherein obtaining a second weight value of the second predicted capacity specifically includes:
and acquiring the second weight value according to the combination of one or more of the number of the application services accessed by the cloud server, the number of the application service users and the application service activity time in a second time period.
6. The capacity adjustment method of the cloud server according to claim 1, further comprising the steps of:
and adjusting the capacity of the cloud server at the time node to be predicted according to the predicted capacity.
7. The capacity adjustment method of the cloud server according to claim 6, further comprising the steps of:
acquiring the actual demand capacity of the cloud server at the time node to be predicted;
obtaining a difference between the actual demand capacity and the predicted capacity;
and if the difference between the actual required capacity and the predicted capacity is larger than a first set threshold, adjusting the capacity of the cloud server according to the difference.
8. The capacity adjustment method of the cloud server according to claim 6, further comprising the steps of:
acquiring the actual demand capacity of the cloud server at the time node to be predicted;
obtaining a difference between the actual demand capacity and the predicted capacity;
if the difference between the actual demand capacity and the predicted capacity is greater than a first set threshold, the difference is sent to an administrator client.
9. The method for adjusting the capacity of the cloud server according to claim 1, wherein a first neural network model is trained by using a time series dataset of the capacity load of the cloud server in the first time interval, and weight and threshold values of the first neural network model are optimized by a horizontal-vertical double-chord optimization algorithm, and the method comprises the following steps:
initializing a first neural network model, and setting an initial weight and a threshold;
initializing an artificial particle population, and carrying out individual coding on the artificial particle population;
training a first neural network model by using the capacity load time series data of the cloud server in a first time interval, and evaluating the fitness value of each individual of the artificial particle population by using a training sample error;
and performing horizontal double-chord global search and vertical double-chord local development on the artificial particle population, and preferentially reserving the generated new individuals and parent individuals.
10. The capacity adjustment method of the cloud server according to claim 9, further comprising the steps of:
and if the error of the training sample is smaller than a second set threshold value or the iteration times of the horizontal and vertical double-chord optimizing algorithm exceed the maximum set iteration times, stopping training the first neural network model, and outputting the current weight and threshold value as the weight and threshold value of the first neural network model.
11. The capacity adjustment method of the cloud server according to claim 1 or 2, wherein the training process of the second neural network model includes:
acquiring a time sequence data set of the cloud server capacity load in a second time interval;
training a second neural network model by using the time series data set of the cloud server capacity load in the second time interval, and optimizing the weight and the threshold of the second neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm.
12. The method for adjusting the capacity of the cloud server according to claim 11, wherein training a second neural network model using the time-series data set of the capacity load of the cloud server in the second time interval, and optimizing the weight and the threshold of the second neural network model by a horizontal-vertical double-chord optimization algorithm comprises:
initializing a second neural network model, and setting an initial weight and a threshold;
initializing an artificial particle population, and carrying out individual coding on the artificial particle population;
training a second neural network model by using the capacity load time series data of the cloud server in a second time interval, and evaluating the fitness value of each individual of the artificial particle population by using the training sample error;
and performing horizontal double-chord global search and vertical double-chord local development on the artificial particle population, and preferentially reserving the generated new individuals and parent individuals.
13. The capacity adjustment method of the cloud server according to claim 12, further comprising the steps of:
and if the error of the training sample is smaller than a third set threshold value or the iteration times of the horizontal and vertical double-chord optimizing algorithm exceed the maximum set iteration times, stopping training the second neural network model, and outputting the current weight and threshold value as the weight and threshold value of the second neural network model.
14. A capacity adjustment device of a cloud server, comprising:
the data acquisition module is used for acquiring time sequence data of the capacity load of the cloud server before the time node to be predicted;
the first input module is used for inputting the time series data into a first neural network model and acquiring first predicted capacity of the cloud server at the time node to be predicted, wherein the first neural network model is trained by the time series data of the capacity load of the cloud server in a first time interval before the time node to be predicted; the training process of the first neural network model comprises the following steps: acquiring a time series data set of the capacity load of the cloud server in a first time interval; training a first neural network model by using a time series dataset of the cloud server capacity load in the first time interval, and optimizing weight and threshold values of the first neural network model through a horizontal and vertical double-chord optimizing algorithm, wherein the horizontal and vertical double-chord optimizing algorithm comprises a horizontal double-chord optimizing algorithm and a vertical double-chord optimizing algorithm;
the second input module is used for inputting the time series data into a second neural network model and acquiring second predicted capacity of the cloud server at the time node to be predicted, wherein the second neural network model is trained by the time series data of the capacity load of the cloud server in a second time interval before the time node to be predicted;
and the prediction module is used for acquiring the predicted capacity of the cloud server at the time node to be predicted according to the first predicted capacity and the second predicted capacity.
15. A computer-readable medium having a computer program stored thereon, characterized in that:
the computer program, when executed by a processor, implements a capacity adjustment method of a cloud server according to any one of claims 1 to 13.
16. A cloud server comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein:
the processor, when executing the computer program, implements the capacity adjustment method of the cloud server according to any one of claims 1 to 13.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910122494.3A CN109992412B (en) | 2019-02-19 | 2019-02-19 | Capacity adjusting method and device of cloud server, storage medium and cloud server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910122494.3A CN109992412B (en) | 2019-02-19 | 2019-02-19 | Capacity adjusting method and device of cloud server, storage medium and cloud server |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109992412A CN109992412A (en) | 2019-07-09 |
| CN109992412B true CN109992412B (en) | 2021-08-10 |
Family
ID=67129841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910122494.3A Active CN109992412B (en) | 2019-02-19 | 2019-02-19 | Capacity adjusting method and device of cloud server, storage medium and cloud server |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109992412B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112328387A (en) * | 2020-09-04 | 2021-02-05 | 广东电力信息科技有限公司 | Multi-tenant-oriented micro-service resource scheduling method |
| US11863453B2 (en) * | 2021-06-01 | 2024-01-02 | Cisco Technology, Inc. | Dynamic load adjustment |
| US11962506B2 (en) | 2021-06-01 | 2024-04-16 | Cisco Technology, Inc. | Dynamic traffic class upgrading |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101695050A (en) * | 2009-10-19 | 2010-04-14 | 浪潮电子信息产业股份有限公司 | Dynamic load balancing method based on self-adapting prediction of network flow |
| CN107040412A (en) * | 2017-03-31 | 2017-08-11 | 南京邮电大学 | A kind of Cloud Server load predicting method |
| CN107045455A (en) * | 2017-06-19 | 2017-08-15 | 华中科技大学 | A kind of Docker Swarm cluster resource method for optimizing scheduling based on load estimation |
| CN107231264A (en) * | 2017-07-25 | 2017-10-03 | 北京百度网讯科技有限公司 | For the method and apparatus for the capacity for managing Cloud Server |
| CN108205698A (en) * | 2017-06-15 | 2018-06-26 | 广东工业大学 | A kind of cloud resource load predicting method based on the double string whale optimization algorithms of just remaining chaos |
| CN109272142A (en) * | 2018-08-14 | 2019-01-25 | 苏州朗动网络科技有限公司 | Data predication method, device, computer equipment and storage medium based on event |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6823380B1 (en) * | 2000-08-07 | 2004-11-23 | Microsoft Corporation | System and method providing continual rate requests |
-
2019
- 2019-02-19 CN CN201910122494.3A patent/CN109992412B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101695050A (en) * | 2009-10-19 | 2010-04-14 | 浪潮电子信息产业股份有限公司 | Dynamic load balancing method based on self-adapting prediction of network flow |
| CN107040412A (en) * | 2017-03-31 | 2017-08-11 | 南京邮电大学 | A kind of Cloud Server load predicting method |
| CN108205698A (en) * | 2017-06-15 | 2018-06-26 | 广东工业大学 | A kind of cloud resource load predicting method based on the double string whale optimization algorithms of just remaining chaos |
| CN107045455A (en) * | 2017-06-19 | 2017-08-15 | 华中科技大学 | A kind of Docker Swarm cluster resource method for optimizing scheduling based on load estimation |
| CN107231264A (en) * | 2017-07-25 | 2017-10-03 | 北京百度网讯科技有限公司 | For the method and apparatus for the capacity for managing Cloud Server |
| CN109272142A (en) * | 2018-08-14 | 2019-01-25 | 苏州朗动网络科技有限公司 | Data predication method, device, computer equipment and storage medium based on event |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109992412A (en) | 2019-07-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11848826B2 (en) | Hyperparameter and network topology selection in network demand forecasting | |
| KR102191444B1 (en) | Training of neural networks using prioritized experiential memory | |
| CN110070117B (en) | Data processing method and device | |
| EP3525096A1 (en) | Resource load balancing control method and cluster scheduler | |
| US20200090073A1 (en) | Method and apparatus for generating machine learning model | |
| CN109992412B (en) | Capacity adjusting method and device of cloud server, storage medium and cloud server | |
| CN110476172A (en) | Neural framework for convolutional neural networks is searched for | |
| US20210027197A1 (en) | Dynamic placement of computation sub-graphs | |
| Kaur et al. | Deep‐Q learning‐based heterogeneous earliest finish time scheduling algorithm for scientific workflows in cloud | |
| Dias et al. | Parallel computing applied to the stochastic dynamic programming for long term operation planning of hydrothermal power systems | |
| US20220076169A1 (en) | Federated machine learning using locality sensitive hashing | |
| CN110795246A (en) | Resource utilization rate prediction method and device | |
| EP3938929A1 (en) | Determining causal models for controlling environments | |
| CN109842563A (en) | Content delivery network flow dispatching method, device and computer readable storage medium | |
| EP3607436B1 (en) | Disaggregating latent causes for computer system optimization | |
| Liu et al. | CORP: Cooperative opportunistic resource provisioning for short-lived jobs in cloud systems | |
| CN111143039B (en) | Scheduling method and device of virtual machine and computer storage medium | |
| Li et al. | An effective scheduling strategy based on hypergraph partition in geographically distributed datacenters | |
| US10506031B1 (en) | Scalable network for processing virtual environments | |
| US11928556B2 (en) | Removing unnecessary history from reinforcement learning state | |
| CN111506414B (en) | Resource scheduling method, device, equipment, system and readable storage medium | |
| Awad et al. | A swarm intelligence-based approach for dynamic data replication in a cloud environment | |
| US20210056457A1 (en) | Hyper-parameter management | |
| CN115700485A (en) | Optimizing deployment of machine learning workloads | |
| CN114741161A (en) | HPC job cluster sensing method based on mixed cluster |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |