CN117560110B - NTP time service method and system for high access request - Google Patents
NTP time service method and system for high access request Download PDFInfo
- Publication number
- CN117560110B CN117560110B CN202410045879.5A CN202410045879A CN117560110B CN 117560110 B CN117560110 B CN 117560110B CN 202410045879 A CN202410045879 A CN 202410045879A CN 117560110 B CN117560110 B CN 117560110B
- Authority
- CN
- China
- Prior art keywords
- time
- adjustment value
- time adjustment
- weight parameter
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000012545 processing Methods 0.000 claims abstract description 26
- 238000004088 simulation Methods 0.000 claims description 121
- 238000012549 training Methods 0.000 claims description 108
- 238000004364 calculation method Methods 0.000 claims description 28
- 210000005036 nerve Anatomy 0.000 claims description 27
- 238000004422 calculation algorithm Methods 0.000 claims description 23
- 230000000007 visual effect Effects 0.000 claims description 19
- 238000005457 optimization Methods 0.000 claims description 16
- 230000004044 response Effects 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 12
- 230000001537 neural effect Effects 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 230000006399 behavior Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 230000001105 regulatory effect Effects 0.000 claims 1
- 230000008859 change Effects 0.000 abstract description 4
- 238000007726 management method Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04J—MULTIPLEX COMMUNICATION
- H04J3/00—Time-division multiplex systems
- H04J3/02—Details
- H04J3/06—Synchronising arrangements
- H04J3/0635—Clock or time synchronisation in a network
- H04J3/0638—Clock or time synchronisation among nodes; Internode synchronisation
- H04J3/0658—Clock or time synchronisation among packet nodes
- H04J3/0661—Clock or time synchronisation among packet nodes using timestamps
- H04J3/0667—Bidirectional timestamps, e.g. NTP or PTP for compensation of clock drift and for compensation of propagation delays
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/04—Generating or distributing clock signals or signals derived directly therefrom
- G06F1/12—Synchronisation of different clock signals provided by a plurality of clock generators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/08—Learning methods
Abstract
The invention relates to the field of time synchronization, in particular to an NTP time service method and system under a high access request. An NTP time service system for use with high access requests, comprising: the system comprises a unit time task processing amount acquisition module, a time service server working state acquisition module, a state judgment module, an NTP time service information acquisition module, a time adjustment value output model management module, a time adjustment value output module, a time adjustment value set storage module, a time adjustment value prediction model management module, a time adjustment value prediction module and an NTP time service module. According to the invention, the local clock of the time synchronization client is time-corrected by the predicted time adjustment value output based on the time sequence change of the time adjustment value, so that NTP time service errors caused by the condition that the time synchronization client is in high task quantity or the time service server is in high access request are avoided, and the precision of NTP time service is further improved.
Description
Technical Field
The invention relates to the field of time synchronization, in particular to an NTP time service method and system under a high access request.
Background
NTP timing services are critical to ensure accuracy and consistency of computer system time. In modern information technology environments, accurate time synchronization is the basis for many critical applications, such as financial transactions, data center operations, network communication protocols, and the like. NTP provides a low cost and efficient solution to calibrate the internal clock of a computer by synchronizing time over the network. This synchronization is particularly critical to maintaining coordination and data integrity between network devices.
However, the accuracy of NTP time service is significantly affected by Round-trip delay (Round-TRIP DELAY) symmetry assumptions. NTP protocols generally assume that the time for a packet to return from a client to a server to the client is the same. This assumption simplifies the computation of clock skew, but in real-world network environments, the round-trip path delay tends to vary due to various factors such as network congestion, router processing speed, different transmission paths, etc. When such symmetry assumptions are not established, the NTP algorithm may not accurately calculate and correct clock skew, thereby affecting the accuracy of clock synchronization.
In addition, the efficiency and accuracy of NTP time service is also affected by the amount of client tasks and the NTP server request load. When the computing resources of a client are occupied by a large number of tasks, it may not be able to handle the response of the NTP service in time, resulting in delays and inaccuracy in time synchronization. Also, if the NTP server faces a high number of access requests, its response time may increase, affecting the quality of service. In extreme cases, this may cause overload of the server, further reducing the accuracy and reliability of the time synchronization.
Disclosure of Invention
According to the invention, the NTP time service information is directly sent into the time adjustment value output model for calculation, so that deep learning of the nonlinear relation between the NTP time service information and the time adjustment value can be realized, the time adjustment value is further output, the direct calculation through various time stamps is not needed, and the problem of time service error caused by directly regarding round trip delay as equal in the traditional NTP time service method is avoided; meanwhile, when the time synchronization client is in high task amount or the time service server is in high access request, the local clock of the time synchronization client is time-corrected through the predicted time adjustment value output based on time sequence change of the time adjustment value, so that NTP time service errors caused when the time synchronization client is in high task amount or the time service server is in high access request are avoided, and the precision of NTP time service is further improved.
An NTP time service method for high access requests, comprising:
Acquiring task processing quantity Q of a unit time of the time synchronization client in preset time before the time synchronization client sends a time synchronization request to the time service server, and acquiring working states of all the time service servers in a mode that the time synchronization client sends a state request message to the time service server, wherein the working states comprise busy and not busy;
Judging the task processing capacity Q of a time synchronization client in unit time and the working states of all time service servers in preset time, if the task processing capacity Q is smaller than G and L/R is smaller than alpha, G is a task processing capacity threshold value in unit time, L is the total number of time service servers with busy working states, R is the total number of time service servers, and alpha is a time service server number threshold value, and executing the step S1; if the Q is not less than G and the L/R is less than alpha, executing the step S2;
S1: the method comprises the steps that a time synchronization client sends a time synchronization request to R time service servers, after any time service server receives NTP time service requests, a corresponding response message is generated and fed back to the time synchronization client, the time synchronization client generates NTP time service information according to the R response messages, the storage form of the NTP time service information is { X 1,X2…Xr…XR }, wherein X r is time service information corresponding to the R-th time service server, the storage form of time service information X r is [ T 1,T2(r),T3(r),T4 (R) ], T 1 is a time stamp of the NTP time service requests sent by the time synchronization client to all the time service servers, T 2 (n) is a time stamp of the NTP time service requests received by the n-th time service server, T 3 (n) is a time stamp of the response messages sent by the n-th time service server to the time synchronization client, and T 4 (n) is a time stamp of the response messages sent by the time synchronization client; the NTP time service information is sent into a trained time adjustment value output model for calculation, the time adjustment value output model is built based on an ELM model improved by a moth flame optimization algorithm, and a corresponding time adjustment value is output; adjusting a local clock of the time synchronization client by the time adjustment value, and sequentially storing the time adjustment value into a time adjustment value set according to the output time sequence;
S2: and outputting a predicted time adjustment value based on the time adjustment value set and the trained time adjustment value prediction model, establishing the time adjustment value prediction model based on the transducer model, adjusting the local clock of the time synchronization client by the predicted time adjustment value, and sequentially storing the predicted time adjustment value into the time adjustment value set according to the output time sequence.
Preferably, the time adjustment value output model comprises a feature extractor and a time adjustment value output device, wherein the feature extractor is built based on the RBM model and comprises a visual layer and an implicit layer, and the number of the neural nodes of the visual layer and the implicit layer is R; the time adjustment value output device is built based on an ELM model improved by a moth flame optimization algorithm and comprises an input layer, a hidden layer and an output layer, wherein the number of the neural nodes of the input layer is R, the number of the neural nodes of the output layer is 1, the number of the neural nodes of the hidden layer is (1+R) 0.5 +beta, and beta is a random integer of 2-10;
And the NTP time service information is subjected to feature learning through a feature extractor, and the time adjustment value is output through a time adjustment value output device.
Preferably, NTP time service information is sent to a trained time adjustment value output model for calculation, and the method specifically comprises the following steps:
Sending time service information X r in the NTP time service information into nerve nodes of a visual layer in a feature extractor one by one, calculating the NTP time service information through a first weight parameter theta 1 between the visual layer and an implicit layer, wherein theta 1＝{Wij(1),ai,bj},Wij is a first connection weight between an ith nerve node in the visual layer and a jth nerve node in the implicit layer, a i is a first bias value corresponding to the ith nerve node in the visual layer, b j is a bias value corresponding to the jth nerve node in the implicit layer, i=1, 2,3 … R, j=1, 2,3 … R, obtaining a feature dataset, and the storage form of the feature dataset is { Y 1,Y2…Yr…YR }, wherein Y r is feature data corresponding to the jth time service information; and sending the characteristic data Y r in the characteristic data set into a time adjustment value output device one by one to calculate, outputting a time adjustment value through an output layer, wherein a second weight parameter theta 2 between the input layer and the hidden layer is obtained through simulation calculation of a moth flame optimization algorithm, theta 2＝{Wku(2),Bu }, wherein W ku (2) is a second connection weight between a kth nerve node in the input layer and a u nerve node in the hidden layer, B u is a second offset value corresponding to the u nerve node in the hidden layer, and k=1, 2,3 … R, and u=1, 2,3 … (1+R) 0.5 +beta.
Preferably, the training of the output model for time adjustment values comprises the steps of:
Acquiring training NTP time service information, dividing all training NTP time service information into a first training set and a first testing set, sending the first training set into a feature extractor of initialization parameters for training, and adjusting a first weight parameter theta 1 by maximizing a logarithmic interpretation function of the feature extractor on the first training set during training; when the maximum training times are reached, outputting a trained feature extractor, and fixing a first weight parameter theta 1 in the feature extractor;
The first training set is sent to a time adjustment value output model initialized by a time adjustment value output device parameter to train, during the period, a first weight parameter theta 1 in a feature extractor is guaranteed to be unchanged, in the training process, a first cross entropy loss value is calculated based on the time adjustment value output by the time adjustment value output model and an actual time adjustment value corresponding to training NTP time service information, whether the first cross entropy loss value is located in a first preset range is judged, if the first cross entropy loss value is located in the first preset range, a trained time adjustment value output model is output; otherwise, continuing to train the time adjustment value output model through the first training set; and adjusting the super parameters of the time adjustment value output model through the first test set.
Preferably, the second weight parameter θ 2 between the input layer and the hidden layer is obtained by simulation calculation of a moth flame optimization algorithm, which specifically comprises the following steps:
S1: establishing N second weight parameter simulation individuals F n, wherein each second weight parameter simulation individual F n is regarded as one moth in a moth flame optimization algorithm, n=1, 2,3 … N, the storage form of the second weight parameter simulation individual F n is { F n,1,fn,2,fn,3…fn,d…fn,D }, F n,d is the D-th simulation parameter value in the second weight parameter simulation individuals F n, D is the serial number corresponding to the simulation parameter value F n,d, d=1, 2,3 … D, D= (1+R) [ (1+R) 0.5 +beta ];
S2: setting the maximum iteration number H, wherein h=1, and H is used for recording the iteration number;
S3: calculating the corresponding fitness delta n of the second weight parameter simulation individuals F n, arranging all the second weight parameter simulation individuals F n according to the corresponding fitness delta n from large to small, forming a population set by all the arranged second weight parameter simulation individuals F n, forming a flame individual set by selecting the first ζ second weight parameter simulation individuals F n, and marking the second weight parameter simulation individuals F n in the flame individual set as flame individuals A μ, μ=1, 2,3 … ζ in an initial state;
S4: selecting second weight parameter simulation individuals F n from the population set one by one, outputting the arrangement sequence number v of the selected second weight parameter simulation individuals F n in the population set aiming at the selected second weight parameter simulation individuals F n, judging whether v > ζ is met, if v > ζ is not met, selecting a v-th flame individual A μ from the flame individual set as an associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n, if v > ζ is met, selecting one flame individual A μ from the flame individual set as an associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n based on the adaptability delta n corresponding to all the second weight parameter simulation individuals F n in the flame individual set and a roulette selection algorithm, and updating the selected second weight parameter simulation individual F n by the following formula based on the associated flame individual F n-con:
fn,d＝dis(Fn,Fn-con)·exp(ct)·cos(2πt)+fn_con,d；
Wherein dis (F n,Fn-con) is the distance between the selected second weight parameter simulated individual F n and the corresponding associated flame individual F n-con; c is the logarithmic spiral shape constant; t is a random number between intervals [ -1,1], F n_con,d is the d-th simulation parameter value in the associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n;
s5: updating the number ζ of flame individuals a n in the set of flame individuals by the following formula:
s6: judging whether H < H is true, if so, entering S7; if "H < H" is not satisfied, entering S8;
S7: calculating the corresponding fitness delta n of the second weight parameter simulation individuals F n, arranging all the second weight parameter simulation individuals F n according to the corresponding fitness delta n from large to small, reorganizing all the arranged second weight parameter simulation individuals F n into a population set, arranging all the second weight parameter simulation individuals F n in the population set and all the second weight parameter simulation individuals F n in the flame individual set according to the corresponding fitness delta n from large to small, and selecting the first ζ second weight parameter simulation individuals F n to form the flame individual set to return to S4;
S8: and selecting all second weight parameter simulation individuals F n in the population set and all second weight parameter simulation individuals F n with the largest fitness delta n in the flame individual set as second weight parameters theta 2 to output.
Preferably, establishing N second weight parameter simulation individuals F n specifically includes the following steps:
S1.1: establishing an intermediate set G with the number of elements being D, and marking the D element in the intermediate set G as G d; selecting elements G d in the middle set G one by one, selecting a random number from a section (f_min d,f_maxd) to assign a value to the elements G d according to the selected elements G d, wherein f_min d is a lower limit value corresponding to the d-th simulation parameter value, and f_max d is an upper limit value corresponding to the d-th simulation parameter value; after all elements G d in the intermediate set G are assigned values, marking the intermediate set G as a second weight parameter simulation individual F n;
S1.2: repeating the step S1.1 for N times, and establishing N second weight parameter simulation individuals F n.
Preferably, calculating the fitness delta n corresponding to the second weight parameter simulation individual F n specifically includes the following steps: the second weight parameter simulation individual F n is used as a second weight parameter theta 2 to be sent into a time adjustment value output model, the time adjustment value output model is tested through a first test set, loss values epsilon e corresponding to the time adjustment value output model in the output test process are e=1, 2,3 … E, E is the total number of all training NTP time service information in the first test set, and the second weight parameter simulation individual F n corresponds to the adaptability
Preferably, the adjustment value prediction model is built based on a transducer model, including an encoder and a decoder;
The adjustment value prediction information is sent to a trained adjustment value prediction model for calculation, and the method specifically comprises the following steps:
And establishing an adjustment value prediction information matrix based on the adjustment value prediction information, wherein the decoder receives the adjustment value prediction information matrix, the size of the adjustment value prediction information matrix is MXV, each behavior time adjustment value in the adjustment value prediction information matrix is converted into a vector representation after the two-level system, and V is the total number of dimensions after each time adjustment value is converted into the two-level system.
Preferably, the training for the adjustment value prediction model comprises the steps of:
Acquiring a training time adjustment value; arranging all training time adjustment values according to the output time, selecting the arranged training time adjustment values through a sliding window with the length of M, forming training samples by the training time adjustment values in the sliding window, and forming a second training set by all training samples; the second training set is sent to an adjustment value prediction model of the initialization parameter for training, in the training process, a second cross entropy loss value is calculated based on a prediction time adjustment value output by the adjustment value prediction model and a training time adjustment value after a sliding window corresponding to a training sample, whether the second cross entropy loss value is located in a second preset range is judged, if the second cross entropy loss value is located in the second preset range, a trained adjustment value prediction model is output; otherwise, continuing to train the adjustment value prediction model through the second training set;
And training the adjustment value prediction model through the time adjustment value set in a preset period, and adjusting training parameters in the adjustment value prediction model.
An NTP time service system for use with high access requests, comprising:
the unit time task processing amount acquisition module is used for acquiring the unit time task processing amount of the time synchronization client in preset time;
The time service server working state acquisition module is used for acquiring the working states of all the time service servers in a mode that the time synchronization client side sends a state request message to the time service server;
The state judging module is used for judging the unit time task processing capacity of the time synchronization client and the working states of all time service servers within the preset time;
The NTP time service information acquisition module is used for acquiring the NTP time service information;
the time adjustment value output model management module is used for storing and training the time adjustment value output model;
The time adjustment value output module is used for sending the NTP time service information into the trained time adjustment value output model for calculation and outputting a corresponding time adjustment value;
The time adjustment value set storage module is used for storing a time adjustment value set, and the time adjustment value set is used for storing the time adjustment value output by the adjustment value output module and the predicted time adjustment value output by the time adjustment value prediction module;
the time adjustment value prediction model management module is used for storing and training a time adjustment value prediction model;
The time adjustment value prediction module is used for sending the time adjustment value set into a trained adjustment value prediction model for calculation and outputting a predicted time adjustment value;
and the NTP time service module is used for adjusting the local clock of the time synchronization client through the time adjustment value or the predicted time adjustment value.
The invention has the following advantages:
1. According to the invention, the NTP time service information is directly sent into the time adjustment value output model for calculation, so that deep learning of the nonlinear relation between the NTP time service information and the time adjustment value can be realized, the time adjustment value is further output, the direct calculation through various time stamps is not needed, and the problem of time service error caused by directly regarding round trip delay as equal in the traditional NTP time service method is avoided; meanwhile, when the time synchronization client is in high task amount or the time service server is in high access request, the local clock of the time synchronization client is time-corrected through the predicted time adjustment value output based on time sequence change of the time adjustment value, so that NTP time service errors caused when the time synchronization client is in high task amount or the time service server is in high access request are avoided, and the precision of NTP time service is further improved.
2. According to the invention, the time adjustment value output model is constructed by combining the RBM model and the ELM model, so that the accuracy of the time adjustment value output model in outputting the time adjustment value is improved, and the accuracy of NTP time service is improved on the premise that the ELM model is fast in data processing.
Drawings
Fig. 1 is a schematic structural diagram of an NTP timing system for use in a high access request according to an embodiment of the present invention.
Detailed Description
In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
Example 1: an NTP time service method for high access requests, comprising:
When the time synchronization client transmits a time synchronization request to the time service server within a preset time, the time service server is an NTP server, time service can be provided based on satellite signals, the preset time can be set in advance by an operator, the time service server can be 5min, the task processing amount Q of the time synchronization client in the unit time within the preset time is obtained by dividing all task processing amounts of the time synchronization client within the preset time by the preset time, meanwhile, the working states of all time service servers are obtained by transmitting state request messages to the time service server through the time synchronization client, the working states comprise busy and not busy, the request messages can be returned to the access request number in an access request queue of the time service server after the time synchronization client transmits the state request messages to the time service server, the judgment of the working states of the time service server can be judged by judging the access request number in the access request queue of the time service server, if the number of the access request number in the access request queue exceeds the expected, the corresponding working states are indicated as busy, and otherwise, the working states are not busy;
Judging the task processing capacity Q of the time synchronization client in unit time and the working states of all time service servers in preset time, if the task processing capacity Q of the time synchronization client is less than G and L/R is less than alpha, the condition that the time synchronization client is not in high task capacity and most of the time service servers are not in high access requests is explained, wherein G is a task processing capacity threshold of the unit time, an operator sets according to the performance of the time synchronization client, L is the total number of the time service servers with busy working states, R is the total number of the time service servers, alpha is a time service server number threshold which is generally set to be 0.7, and executing step S1; if the Q is not less than G and the L/R is less than alpha, executing the step S2;
S1: the method comprises the steps that a time synchronization client sends a time synchronization request to R time service servers, after any time service server receives NTP time service requests, a corresponding response message is generated, the response message comprises timestamp information, the response message is fed back to the time synchronization client, the time synchronization client generates NTP time service information according to the R response messages, the storage form of the NTP time service information is { X 1,X2…Xr…XR }, wherein X r is time service information corresponding to the R time service server, the storage form of time service information X r is [ T 1,T2(r),T3(r),T4 (R) ], T 1 is a timestamp of the NTP time service requests sent to all time service servers by the time synchronization client, T 2 (n) is a timestamp of the NTP time service requests received by the n time service server, T 3 (n) is a timestamp of the response messages sent to the time synchronization client by the n time service server, and T 4 (n) is a timestamp of the response messages sent by the n time synchronization client; the NTP time service information is sent into a trained time adjustment value output model for calculation, the time adjustment value output model is built based on an ELM model improved by a moth flame optimization algorithm, and a corresponding time adjustment value is output; adjusting a local clock of the time synchronization client by the time adjustment value, and sequentially storing the time adjustment value into a time adjustment value set according to the output time sequence;
The time adjustment value output model comprises a feature extractor and a time adjustment value output device, wherein the feature extractor is built based on the RBM model and comprises a visual layer and an implicit layer, and the number of the neural nodes of the visual layer and the implicit layer is R, so that NTP time service information is conveniently processed; the time adjustment value output device is built based on an ELM model improved by a moth flame optimization algorithm and comprises an input layer, a hidden layer and an output layer, specifically, parameters between the input layer and the hidden layer are improved by the moth flame optimization algorithm, the number of nerve nodes of the input layer is R, the number of nerve nodes of the output layer is 1, the number of nerve nodes of the hidden layer is (1+R) 0.5 +beta, and beta is a random integer of 2-10, and in actual training, beta can be regarded as a super parameter to be adjusted and set; it should be noted that, the internal structures of the feature extractor and the time adjustment value output device are respectively set with reference to the traditional RBM model and ELM model;
The NTP time service information is subjected to feature learning through a feature extractor, and a time adjustment value is output through a time adjustment value output device;
the NTP time service information is sent into a trained time adjustment value output model for calculation, and the method specifically comprises the following steps:
The method comprises the steps that time service information X r in NTP time service information is sent to nerve nodes of a visual layer in a feature extractor one by one, the NTP time service information is calculated through a first weight parameter theta 1 between the visual layer and an implicit layer, theta 1＝{Wij(1),ai,bj},Wij is first connection weight between an ith nerve node in the visual layer and a jth nerve node in the implicit layer, a i is a first bias value corresponding to the ith nerve node in the visual layer, b j is a bias value corresponding to the jth nerve node in the implicit layer, i=1, 2,3 … R, j=1, 2,3 … R, a feature data set is obtained, the feature data set can represent deep features corresponding to the NTP time service information, the follow-up ELM model can be helped to further learn a nonlinear relation between the NTP time service information and an output time adjustment value, a storage form of the feature data set is { Y 1,Y2…Yr…YR }, wherein Y r is feature data corresponding to the ith time service information in the visual layer, and can be in a predicted feature data storage form consistent with the time service information set in the NTP time service form; and sending the characteristic data Y r in the characteristic data set into a time adjustment value output device one by one to calculate, outputting a time adjustment value through an output layer, wherein a second weight parameter theta 2 between the input layer and the hidden layer is obtained through simulation calculation of a moth flame optimization algorithm, theta 2＝{Wku(2),Bu }, wherein W ku (2) is a second connection weight between a kth nerve node in the input layer and a u nerve node in the hidden layer, B u is a second offset value corresponding to the u nerve node in the hidden layer, and k=1, 2,3 … R, and u=1, 2,3 … (1+R) 0.5 +beta.
According to the application, the time adjustment value output model is constructed by combining the RBM model and the ELM model, so that the accuracy of the time adjustment value output model in outputting the time adjustment value is improved, and the accuracy of NTP time service is improved on the premise that the ELM model is fast in data processing.
S2: and outputting a predicted time adjustment value based on the time adjustment value set and the trained time adjustment value prediction model, establishing the time adjustment value prediction model based on the transducer model, adjusting the local clock of the time synchronization client by the predicted time adjustment value, and sequentially storing the predicted time adjustment value into the time adjustment value set according to the output time sequence.
According to the application, the NTP time service information is directly sent into the time adjustment value output model for calculation, so that deep learning of the nonlinear relation between the NTP time service information and the time adjustment value can be realized, the time adjustment value is further output, the direct calculation through various time stamps is not needed, and the problem of time service error caused by directly regarding round trip delay as equal in the traditional NTP time service method is avoided; meanwhile, when the time synchronization client is in high task amount or the time service server is in high access request, the local clock of the time synchronization client is time-corrected through the predicted time adjustment value output based on time sequence change of the time adjustment value, so that NTP time service errors caused when the time synchronization client is in high task amount or the time service server is in high access request are avoided, and the precision of NTP time service is further improved.
Training of the output model for time adjustment values comprises the following steps:
Acquiring training NTP time service information, dividing all training NTP time service information into a first training set and a first testing set, wherein the dividing ratio can be 8:2, sending the first training set to a feature extractor of the initialization parameter for training, wherein the training mode can adopt a CD algorithm, and the first weight parameter theta 1 is adjusted by maximizing a logarithmic interpretation function of the feature extractor on the first training set during the training; when the maximum training times are reached, outputting a trained feature extractor, and fixing a first weight parameter theta 1 in the feature extractor;
Marking all training NTP time service information in a first training set through an actual time adjustment value, wherein the actual time adjustment value is obtained by manually calculating according to standard time and time corresponding to a time synchronization client, sending the first training set into a time adjustment value output model initialized by a time adjustment value output device parameter for training, ensuring that a first weight parameter theta 1 in a feature extractor is unchanged during training, calculating a first cross entropy loss value based on the time adjustment value output by the time adjustment value output model and the actual time adjustment value corresponding to training NTP time service information during training, and judging whether the first cross entropy loss value is in a first preset range or not, and outputting a trained time adjustment value output model if the first cross entropy loss value is in the first preset range; otherwise, continuing to train the time adjustment value output model through the first training set; and adjusting the super parameters of the time adjustment value output model through the first test set.
The second weight parameter theta 2 between the input layer and the hidden layer is obtained through simulation calculation of the moth flame optimization algorithm, and the method specifically comprises the following steps:
S1: establishing N second weight parameter simulation individuals F n, wherein each second weight parameter simulation individual F n is regarded as a moth in a moth flame optimization algorithm, n=1, 2,3 … N, the storage form of the second weight parameter simulation individual F n is { F n,1,fn,2,fn,3…fn,d…fn,D }, F n,d is the D-th simulation parameter value in the second weight parameter simulation individual F n, D is the serial number corresponding to the simulation parameter value F n,d, d=1, 2,3 … D, d= (1+r) [ (1+r) 0.5 +β ], and it is to be noted that, in the second weight parameter simulation individual F n, the former R [ (1+r) 0.5 +β ] simulation parameter values are corresponding second connection weights W ku (2), such as { W 11(2),W12(2),W13(2)…W21 (2) … }, and the latter R simulation parameter values are corresponding second bias values B u;
Establishing N second weight parameter simulation individuals F n, which specifically comprise the following steps:
S1.1: establishing an intermediate set G with the number of elements being D, and marking the D element in the intermediate set G as G d; selecting elements G d in the middle set G one by one, selecting a random number from a section (f_min d,f_maxd) to assign a value to the elements G d according to the selected elements G d, wherein f_min d is a lower limit value corresponding to the d-th simulation parameter value, and f_max d is an upper limit value corresponding to the d-th simulation parameter value; after all elements G d in the intermediate set G are assigned values, marking the intermediate set G as a second weight parameter simulation individual F n;
S1.2: repeating the step S1.1 for N times, and establishing N second weight parameter simulation individuals F n.
S2: setting the maximum iteration number H, wherein h=1, and H is used for recording the iteration number;
S3: calculating the corresponding fitness delta n of the second weight parameter simulation individuals F n, arranging all the second weight parameter simulation individuals F n according to the corresponding fitness delta n from large to small, forming a population set by all the arranged second weight parameter simulation individuals F n, forming a flame individual set by selecting the first ζ second weight parameter simulation individuals F n, and marking the second weight parameter simulation individuals F n in the flame individual set as flame individuals A μ, μ=1, 2,3 … ζ in an initial state;
The calculating of the fitness delta n corresponding to the second weight parameter simulation individual F n specifically includes the following steps: the second weight parameter simulation individual F n is used as a second weight parameter theta 2 to be sent into a time adjustment value output model, the time adjustment value output model is tested through a first test set, loss values epsilon e corresponding to the time adjustment value output model in the output test process are e=1, 2,3 … E, E is the total number of all training NTP time service information in the first test set, and the second weight parameter simulation individual F n corresponds to the adaptability The confidence that the second weight parameter simulation individual F n is taken as the second weight parameter theta 2 is measured through the accuracy calculated by the first test set passing time adjustment value output model, and the higher the confidence is, the higher the corresponding adaptability is;
S4: selecting second weight parameter simulation individuals F n from the population set one by one, outputting an arrangement sequence number v of the selected second weight parameter simulation individuals F n in the population set for the selected second weight parameter simulation individuals F n, judging whether v > ζ 'is satisfied, wherein in an initial state, the arrangement sequence number v cannot be larger than ζ, but ζ decreases along with the increase of the iteration number, so that the situation that the arrangement sequence number v is larger than ζ occurs, if v > ζ' is not satisfied, selecting a v-th flame individual A μ from the flame individual set as an associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n, if v > ζ is satisfied, selecting one flame individual A μ from the flame individual set as an associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n based on the suitability delta n corresponding to all the second weight parameter simulation individuals F n in the flame individual set and a roulette selection algorithm, updating the selected second weight parameter simulation individual F n based on the associated flame individual F n-con by the following formula:
fn,d＝dis(Fn,Fn-con)·exp(ct)·cos(2πt)+fn_con,d；
Wherein dis (F n,Fn-con) is the distance between the selected second weight parameter simulation individual F n and the corresponding associated flame individual F n-con, which can be calculated by a cosine distance algorithm; c is the logarithmic spiral shape constant; t is a random number between intervals [ -1,1], F n_con,d is the d-th simulation parameter value in the associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n;
It should be noted that, in the process of updating the second weight parameter simulation individual F n, if the updated simulation parameter value exceeds the corresponding range (f_min d,f_maxd), the updating may be performed again, or the adjustment may be performed through normalization;
s5: updating the number ζ of flame individuals a μ in the set of flame individuals by the following formula:
the convergence speed of the algorithm can be improved through the reduction of the number of flame individuals A a, and the utilization of the optimal solution is improved;
s6: judging whether 'H < H' is established, if so, indicating that the maximum iteration number is not reached, and entering S7; if 'H < H' is not established, indicating that the maximum iteration number is reached, and entering S8;
S7: calculating the corresponding fitness delta n of the second weight parameter simulation individuals F n, arranging all the second weight parameter simulation individuals F n according to the corresponding fitness delta n from large to small, reorganizing all the arranged second weight parameter simulation individuals F n into a population set, arranging all the second weight parameter simulation individuals F n in the population set and all the second weight parameter simulation individuals F n in the flame individual set according to the corresponding fitness delta n from large to small, and selecting the first ζ second weight parameter simulation individuals F n to form the flame individual set to return to S4;
S8: and selecting all second weight parameter simulation individuals F n in the population set and all second weight parameter simulation individuals F n with the largest fitness delta n in the flame individual set as second weight parameters theta 2 to output.
The predicted time adjustment value is obtained by:
M elements after the time adjustment value set are selected to form adjustment value prediction information; it should be noted that, the adjustment value prediction information needs M time adjustment values to form the adjustment value prediction information, so in actual operation, when the time synchronization client is not busy, M time adjustment values can be output through the time adjustment value output model, and then the prediction time adjustment value is obtained;
And sending the adjustment value prediction information into a trained adjustment value prediction model for calculation, and outputting a prediction time adjustment value.
The adjustment value prediction model is built based on a transducer model, and comprises an encoder and a decoder;
The adjustment value prediction information is sent to a trained adjustment value prediction model for calculation, and the method specifically comprises the following steps:
The method comprises the steps that an adjustment value prediction information matrix is established based on adjustment value prediction information, a decoder receives the adjustment value prediction information matrix, the size of the adjustment value prediction information matrix is MxV, each behavior time adjustment value in the adjustment value prediction information matrix is converted into a vector representation after two-level system, V is the total number of dimensions after each time adjustment value is converted into the two-level system, it is to be noted that V is set by an operator according to the length range after the time adjustment value is converted into the two-level system, and when the time adjustment value is not reached to V after the time adjustment value is converted into the two-level system, the header is filled with 0;
Training for the adjustment value prediction model comprises the following steps:
Acquiring a training time adjustment value, wherein the training time adjustment value refers to a time adjustment value output by a time adjustment value output model in a history record; arranging all training time adjustment values according to output time, selecting the arranged training time adjustment values through a sliding window with the length of M, namely selecting one training time adjustment value from all the arranged training time adjustment values as a starting point, selecting M-1 training time adjustment values backwards based on the starting point, taking the training time adjustment value corresponding to the starting point and the M-1 training time adjustment values selected backwards as training time adjustment values selected by the sliding window, forming training samples by the training time adjustment values in the sliding window, and forming a second training set by all the training samples; the second training set is sent to an adjustment value prediction model of the initialization parameter for training, in the training process, a second cross entropy loss value is calculated based on a prediction time adjustment value output by the adjustment value prediction model and a training time adjustment value after a sliding window corresponding to a training sample, whether the second cross entropy loss value is located in a second preset range or not is judged, the second preset range is set by an operator according to experience, and if the second cross entropy loss value is located in the second preset range, the trained adjustment value prediction model is output; otherwise, continuing to train the adjustment value prediction model through the second training set.
And in the preset period, the preset period is a period of time for which the synchronous client is idle, for example, 00:00-4:00, the adjustment value prediction model is trained through the time adjustment value set, and training parameters in the adjustment value prediction model are adjusted, so that the adjustment value prediction model can be closer to the actual situation.
Example 2: an NTP time service system for use with high access requests, see fig. 1, comprising:
the unit time task processing amount acquisition module is used for acquiring the unit time task processing amount of the time synchronization client in preset time;
The time service server working state acquisition module is used for acquiring the working states of all the time service servers in a mode that the time synchronization client side sends a state request message to the time service server;
The state judging module is used for judging the unit time task processing capacity of the time synchronization client and the working states of all time service servers within the preset time;
The NTP time service information acquisition module is used for acquiring the NTP time service information;
the time adjustment value output model management module is used for storing and training the time adjustment value output model;
The time adjustment value output module is used for sending the NTP time service information into the trained time adjustment value output model for calculation and outputting a corresponding time adjustment value;
The time adjustment value set storage module is used for storing a time adjustment value set, and the time adjustment value set is used for storing the time adjustment value output by the adjustment value output module and the predicted time adjustment value output by the time adjustment value prediction module;
the time adjustment value prediction model management module is used for storing and training a time adjustment value prediction model;
The time adjustment value prediction module is used for sending the time adjustment value set into a trained adjustment value prediction model for calculation and outputting a predicted time adjustment value;
and the NTP time service module is used for adjusting the local clock of the time synchronization client through the time adjustment value or the predicted time adjustment value.
It will be understood that modifications and variations will be apparent to those skilled in the art from the foregoing description, and it is intended that all such modifications and variations be included within the scope of the following claims. Parts of the specification not described in detail belong to the prior art known to those skilled in the art.
Claims (4)
1. An NTP time service method for use in a high access request, comprising:
Acquiring task processing quantity Q of a unit time of the time synchronization client in preset time before the time synchronization client sends a time synchronization request to the time service server, and acquiring working states of all the time service servers in a mode that the time synchronization client sends a state request message to the time service server, wherein the working states comprise busy and not busy; judging the task processing capacity Q of a time synchronization client in unit time and the working states of all time service servers in preset time, if the task processing capacity Q is smaller than G and L/R is smaller than alpha, G is a task processing capacity threshold value in unit time, L is the total number of time service servers with busy working states, R is the total number of time service servers, and alpha is a time service server number threshold value, and executing the step S1; if the Q is not less than G and the L/R is less than alpha, executing the step S2;
S1: the method comprises the steps that a time synchronization client sends a time synchronization request to R time service servers, after any time service server receives NTP time service requests, a corresponding response message is generated and fed back to the time synchronization client, the time synchronization client generates NTP time service information according to the R response messages, the storage form of the NTP time service information is { X 1,X2…Xr…XR }, wherein X r is time service information corresponding to the R-th time service server, the storage form of time service information X r is [ T 1,T2(r),T3(r),T4 (R) ], T 1 is a time stamp of the NTP time service requests sent by the time synchronization client to all the time service servers, T 2 (n) is a time stamp of the NTP time service requests received by the n-th time service server, T 3 (n) is a time stamp of the response messages sent by the n-th time service server to the time synchronization client, and T 4 (n) is a time stamp of the response messages sent by the time synchronization client; the NTP time service information is sent into a trained time adjustment value output model for calculation, the time adjustment value output model is built based on an ELM model improved by a moth flame optimization algorithm, and a corresponding time adjustment value is output; adjusting a local clock of the time synchronization client by the time adjustment value, and sequentially storing the time adjustment value into a time adjustment value set according to the output time sequence;
S2: outputting a predicted time adjustment value based on the time adjustment value set and a trained time adjustment value prediction model, establishing the time adjustment value prediction model based on a transducer model, adjusting a local clock of the time synchronization client by the predicted time adjustment value, and sequentially storing the predicted time adjustment value into the time adjustment value set according to an output time sequence; the time adjustment value output model comprises a feature extractor and a time adjustment value output device, wherein the feature extractor is built based on the RBM model and comprises a visual layer and an implicit layer, and the number of the neural nodes of the visual layer and the implicit layer is R; the time adjustment value output device is built based on an ELM model improved by a moth flame optimization algorithm and comprises an input layer, a hidden layer and an output layer, wherein the number of the neural nodes of the input layer is R, the number of the neural nodes of the output layer is 1, the number of the neural nodes of the hidden layer is (1+R) 0.5 +beta, and beta is a random integer of 2-10;
The NTP time service information is subjected to feature learning through a feature extractor, and a time adjustment value is output through a time adjustment value output device;
the NTP time service information is sent into a trained time adjustment value output model for calculation, and the method specifically comprises the following steps:
Sending time service information X r in the NTP time service information into nerve nodes of a visual layer in a feature extractor one by one, calculating the NTP time service information through a first weight parameter theta 1 between the visual layer and an implicit layer, wherein theta 1＝{Wij(1),ai,bj},Wij is a first connection weight between an ith nerve node in the visual layer and a jth nerve node in the implicit layer, a i is a first bias value corresponding to the ith nerve node in the visual layer, b j is a bias value corresponding to the jth nerve node in the implicit layer, i=1, 2,3 … R, j=1, 2,3 … R, obtaining a feature dataset, and the storage form of the feature dataset is { Y 1,Y2…Yr…YR }, wherein Y r is feature data corresponding to the jth time service information; the characteristic data Y r in the characteristic data set is sent into a time adjustment value output device one by one to be calculated, a time adjustment value is output through an output layer, wherein a second weight parameter theta 2 between an input layer and a hidden layer is obtained through simulation calculation of a moth flame optimization algorithm, theta 2＝{Wku(2),Bu }, W ku (2) is a second connection weight between a kth nerve node in the input layer and a u nerve node in the hidden layer, B u is a second offset value corresponding to the u nerve node in the hidden layer, k=1, 2,3 … R, and u=1, 2,3 … (1+R) 0.5 +beta;
training of the output model for time adjustment values comprises the following steps:
Acquiring training NTP time service information, dividing all training NTP time service information into a first training set and a first testing set, sending the first training set into a feature extractor of initialization parameters for training, and adjusting a first weight parameter theta 1 by maximizing a logarithmic interpretation function of the feature extractor on the first training set during training; when the maximum training times are reached, outputting a trained feature extractor, and fixing a first weight parameter theta 1 in the feature extractor;
The first training set is sent to a time adjustment value output model initialized by a time adjustment value output device parameter to train, during the period, a first weight parameter theta 1 in a feature extractor is guaranteed to be unchanged, in the training process, a first cross entropy loss value is calculated based on the time adjustment value output by the time adjustment value output model and an actual time adjustment value corresponding to training NTP time service information, whether the first cross entropy loss value is located in a first preset range is judged, if the first cross entropy loss value is located in the first preset range, a trained time adjustment value output model is output; otherwise, continuing to train the time adjustment value output model through the first training set; adjusting the super parameters of the time adjustment value output model through the first test set;
the second weight parameter theta 2 between the input layer and the hidden layer is obtained through simulation calculation of the moth flame optimization algorithm, and the method specifically comprises the following steps:
S1: establishing N second weight parameter simulation individuals F n, wherein each second weight parameter simulation individual F n is regarded as one moth in a moth flame optimization algorithm, n=1, 2,3 … N, the storage form of the second weight parameter simulation individual F n is { F n,1,fn,2,fn,3…fn,d…fn,D }, F n,d is the D-th simulation parameter value in the second weight parameter simulation individuals F n, D is the serial number corresponding to the simulation parameter value F n,d, d=1, 2,3 … D, D= (1+R) [ (1+R) 0.5 +beta ];
S2: setting the maximum iteration number H, wherein h=1, and H is used for recording the iteration number;
S3: calculating the corresponding fitness delta n of the second weight parameter simulation individuals F n, arranging all the second weight parameter simulation individuals F n according to the corresponding fitness delta n from large to small, forming a population set by all the arranged second weight parameter simulation individuals F n, forming a flame individual set by selecting the first ζ second weight parameter simulation individuals F n, and marking the second weight parameter simulation individuals F n in the flame individual set as flame individuals A μ, μ=1, 2,3 … ζ in an initial state;
S4: selecting second weight parameter simulation individuals F n from the population set one by one, outputting the arrangement sequence number v of the selected second weight parameter simulation individuals F n in the population set aiming at the selected second weight parameter simulation individuals F n, judging whether v > ζ is met, if v > ζ is not met, selecting a v-th flame individual A μ from the flame individual set as an associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n, if v > ζ is met, selecting one flame individual A μ from the flame individual set as an associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n based on the adaptability delta n corresponding to all the second weight parameter simulation individuals F n in the flame individual set and a roulette selection algorithm, and updating the selected second weight parameter simulation individual F n by the following formula based on the associated flame individual F n-con:
fn,d＝dis(Fn,Fn-con)·exp(ct)·cos(2πt)+fn_con,d；
Wherein dis (F n,Fn-con) is the distance between the selected second weight parameter simulated individual F n and the corresponding associated flame individual F n-con; c is the logarithmic spiral shape constant; t is a random number between intervals [ -1,1], F n_con,d is the d-th simulation parameter value in the associated flame individual F n_con corresponding to the selected second weight parameter simulation individual F n;
s5: updating the number ζ of flame individuals a μ in the set of flame individuals by the following formula:
s6: judging whether H < H is true, if so, entering S7; if "H < H" is not satisfied, entering S8;
S7: calculating the corresponding fitness delta n of the second weight parameter simulation individuals F n, arranging all the second weight parameter simulation individuals F n according to the corresponding fitness delta n from large to small, reorganizing all the arranged second weight parameter simulation individuals F n into a population set, arranging all the second weight parameter simulation individuals F n in the population set and all the second weight parameter simulation individuals F n in the flame individual set according to the corresponding fitness delta n from large to small, and selecting the first ζ second weight parameter simulation individuals F n to form the flame individual set to return to S4;
S8: selecting all second weight parameter simulation individuals F n in the population set and all second weight parameter simulation individuals F n with the largest fitness delta n in the flame individual set as second weight parameters theta 2 to output;
Establishing N second weight parameter simulation individuals F n, which specifically comprise the following steps:
S1.1: establishing an intermediate set G with the number of elements being D, and marking the D element in the intermediate set G as G d; selecting elements G d in the middle set G one by one, selecting a random number from a section (f_min d,f_maxd) to assign a value to the elements G d according to the selected elements G d, wherein f_min d is a lower limit value corresponding to the d-th simulation parameter value, and f_max d is an upper limit value corresponding to the d-th simulation parameter value; after all elements G d in the intermediate set G are assigned values, marking the intermediate set G as a second weight parameter simulation individual F n;
s1.2: repeating the step S1.1 for N times, and establishing N second weight parameter simulation individuals F n;
The calculating of the fitness delta n corresponding to the second weight parameter simulation individual F n specifically includes the following steps: the second weight parameter simulation individual F n is used as a second weight parameter theta 2 to be sent into a time adjustment value output model, the time adjustment value output model is tested through a first test set, loss values epsilon e corresponding to the time adjustment value output model in the output test process are e=1, 2,3 … E, E is the total number of all training NTP time service information in the first test set, and the second weight parameter simulation individual F n corresponds to the adaptability
2. The NTP timing method according to claim 1 wherein the adjustment value prediction model is built based on a transducer model, including an encoder and a decoder;
The adjustment value prediction information is sent to a trained adjustment value prediction model for calculation, and the method specifically comprises the following steps:
And establishing an adjustment value prediction information matrix based on the adjustment value prediction information, wherein the decoder receives the adjustment value prediction information matrix, the size of the adjustment value prediction information matrix is MXV, each behavior time adjustment value in the adjustment value prediction information matrix is converted into a vector representation after the two-level system, and V is the total number of dimensions after each time adjustment value is converted into the two-level system.
3. The NTP timing method for use in high access requests according to claim 2, wherein training for the regulatory value predictive model comprises the steps of:
Acquiring a training time adjustment value; arranging all training time adjustment values according to the output time, selecting the arranged training time adjustment values through a sliding window with the length of M, forming training samples by the training time adjustment values in the sliding window, and forming a second training set by all training samples; the second training set is sent to an adjustment value prediction model of the initialization parameter for training, in the training process, a second cross entropy loss value is calculated based on a prediction time adjustment value output by the adjustment value prediction model and a training time adjustment value after a sliding window corresponding to a training sample, whether the second cross entropy loss value is located in a second preset range is judged, if the second cross entropy loss value is located in the second preset range, a trained adjustment value prediction model is output; otherwise, continuing to train the adjustment value prediction model through the second training set;
And training the adjustment value prediction model through the time adjustment value set in a preset period, and adjusting training parameters in the adjustment value prediction model.
4. An NTP timing system for use in a high access request, wherein the system is applied to an NTP timing method for use in a high access request as claimed in any one of claims 1 to 3, and the system comprises:
the unit time task processing amount acquisition module is used for acquiring the unit time task processing amount of the time synchronization client in preset time;
The time service server working state acquisition module is used for acquiring the working states of all the time service servers in a mode that the time synchronization client side sends a state request message to the time service server;
The state judging module is used for judging the unit time task processing capacity of the time synchronization client and the working states of all time service servers within the preset time;
The NTP time service information acquisition module is used for acquiring the NTP time service information;
the time adjustment value output model management module is used for storing and training the time adjustment value output model;
The time adjustment value output module is used for sending the NTP time service information into the trained time adjustment value output model for calculation and outputting a corresponding time adjustment value;
The time adjustment value set storage module is used for storing a time adjustment value set, and the time adjustment value set is used for storing the time adjustment value output by the adjustment value output module and the predicted time adjustment value output by the time adjustment value prediction module;
the time adjustment value prediction model management module is used for storing and training a time adjustment value prediction model;
The time adjustment value prediction module is used for sending the time adjustment value set into a trained adjustment value prediction model for calculation and outputting a predicted time adjustment value;
and the NTP time service module is used for adjusting the local clock of the time synchronization client through the time adjustment value or the predicted time adjustment value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410045879.5A CN117560110B (en) | 2024-01-12 | 2024-01-12 | NTP time service method and system for high access request |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410045879.5A CN117560110B (en) | 2024-01-12 | 2024-01-12 | NTP time service method and system for high access request |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117560110A CN117560110A (en) | 2024-02-13 |
| CN117560110B true CN117560110B (en) | 2024-04-26 |
Family
ID=89820923
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410045879.5A Active CN117560110B (en) | 2024-01-12 | 2024-01-12 | NTP time service method and system for high access request |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117560110B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108322281A (en) * | 2018-05-14 | 2018-07-24 | 上海市计量测试技术研究院 | Time calibration in network service based on NTP and assessment system and method |
| CN112187560A (en) * | 2020-07-22 | 2021-01-05 | 中国计量科学研究院 | NTP network time service evaluation system, method and time service evaluation device |
| CN115208806A (en) * | 2022-07-07 | 2022-10-18 | 电信科学技术第五研究所有限公司 | Method and device for testing response capability of NTP (network time protocol) server |
| CN116566533A (en) * | 2023-05-24 | 2023-08-08 | 北京无线电计量测试研究所 | Nanosecond NTP network time synchronization method and system |
-
2024
- 2024-01-12 CN CN202410045879.5A patent/CN117560110B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108322281A (en) * | 2018-05-14 | 2018-07-24 | 上海市计量测试技术研究院 | Time calibration in network service based on NTP and assessment system and method |
| CN112187560A (en) * | 2020-07-22 | 2021-01-05 | 中国计量科学研究院 | NTP network time service evaluation system, method and time service evaluation device |
| CN115208806A (en) * | 2022-07-07 | 2022-10-18 | 电信科学技术第五研究所有限公司 | Method and device for testing response capability of NTP (network time protocol) server |
| CN116566533A (en) * | 2023-05-24 | 2023-08-08 | 北京无线电计量测试研究所 | Nanosecond NTP network time synchronization method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117560110A (en) | 2024-02-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021057245A1 (en) | Bandwidth prediction method and apparatus, electronic device and storage medium | |
| CN112465138A (en) | Model distillation method, device, storage medium and equipment | |
| US20100185781A1 (en) | System and Method for Measuring Clock Skew on a Network | |
| CN108647329A (en) | Processing method, device and the computer readable storage medium of user behavior data | |
| CN117560110B (en) | NTP time service method and system for high access request | |
| CN105306553A (en) | Access request scheduling method and device | |
| CN110113660B (en) | Method, device, terminal and storage medium for transcoding time length estimation | |
| CN111680083B (en) | Intelligent multi-level government financial data acquisition system and data acquisition method | |
| CN110543509B (en) | Monitoring system, method and device for user access data and electronic equipment | |
| CN112771816A (en) | Method and device for predicting network rate | |
| CN112383443B (en) | Parallel application communication performance prediction method running in RDMA communication environment | |
| KR102311412B1 (en) | Parameter estimation method based on error term independence correction, and its program implementation and performance system | |
| CN111092755B (en) | Edge service migration simulation method based on resource occupation | |
| CN102982231A (en) | Quantitative calculation method for software confidence level | |
| Orlov et al. | Approach to estimation of performance measures for SIP server model with batch arrivals | |
| CN112671493B (en) | Time transmission quality evaluation method in complex network environment | |
| Yamakawa et al. | Worst-case design of structures using stopping rules in k-adaptive random sampling approach | |
| CN111082973A (en) | Network delay estimation method based on lever sampling | |
| Begin et al. | Towards an automatic modeling tool for observed system behavior | |
| Guan et al. | Performance analysis of polling‐based MAC protocol with retrial for Internet of Things | |
| CN108599834A (en) | A kind of satellite communication network link utilization analysis method and system | |
| CN111104569A (en) | Region segmentation method and device for database table and storage medium | |
| CN113342744B (en) | Parallel construction method, device and equipment of call chain and storage medium | |
| CN106961365A (en) | A kind of network delay measuring method based on Transmission Control Protocol | |
| Chen et al. | An M/G/1 Retrial Queue with Priority, Balking and Feedback Customers. |
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 |