Network parameter configuration method, device, computer equipment and storage medium

Abstract

The present application discloses a network parameter configuration method, device, computer equipment and storage medium, which belongs to the field of data transmission technology. The method inputs the network operation data corresponding to the first time period into the prediction model, so that the prediction model predicts the value of the specified parameter of the network device in the second time period based on the input network operation data, and configures the specified parameter of the network device in the second time period to the value predicted by the prediction model. In this configuration method, there is no need to consider the application scenario, nor to manually modify or configure the parameters, so this configuration method is universal.

Description

[0001] This application is a divisional application. The application number of the original application is 201910878276.2, and the original application date is September 17, 2019. The entire contents of the original application are incorporated into this application by reference. Technical Field

[0002] The present application relates to the technical field of data transmission, and in particular to a network parameter configuration method, device, computer equipment and storage medium. Background Art

[0003] With the widespread application of data centers, data centers require switching networks to have characteristics such as high throughput and low latency. Switches in switching networks can reasonably configure network parameters and regulate the forwarding of messages according to the configured network parameters to avoid network congestion when switches forward messages, thereby enabling switching networks to have characteristics such as high throughput and low latency.

[0004] At present, the process of configuring network parameters may be: taking the network parameters as explicit congestion notification (ECN) parameters as an example, the switch may configure the ECN parameters according to the configuration formula, wherein the configuration formula may be: ECN parameter = Cout*Target_Rtt, Cout is the throughput of the congested port of the switch, Target_Rtt is the end-to-end target delay, so that the switch may directly obtain the ECN parameters according to the configuration formula, and regulate the forwarding of the message according to the obtained ECN parameters to avoid network congestion.

[0005] Since Target_Rtt in the above configuration formula is a static parameter and the value of Target_Rtt is affected by the packet transmission situation, in some application scenarios, the ECN parameter obtained by the switch based on the configuration formula may not be the optimal ECN parameter in the application scenario. Therefore, this configuration method is not universal. Summary of the invention

[0006] The embodiments of the present application provide a network parameter configuration method, apparatus, computer equipment, and storage medium to improve the universality of the network parameter configuration method.

[0007] In a first aspect, a network parameter configuration method is provided, the method comprising:

[0008] Acquire network operation data corresponding to a first time period, the network operation data including at least one of a network transmission characteristic of a network device in the first time period and a first value corresponding to a specified parameter;

[0009] inputting the network operation data into a prediction model;

[0010] Obtaining a second value of the specified parameter output by the prediction model based on the network operation data;

[0011] The designated parameter of the network device in a second time period is configured as the second value, and the second time period is a time period after the first time period.

[0012] This method inputs the network operation data corresponding to the first time period into the prediction model, so that the prediction model predicts the value of the specified parameter of the network device in the second time period based on the input network operation data, and configures the specified parameter of the network device in the second time period to the value predicted by the prediction model. In this configuration method, there is no need to consider the application scenario, and there is no need to manually modify or configure the parameters, so this configuration method is universal.

[0013] In one possible implementation, the method includes:

[0014] Acquire network operation data corresponding to a first time period, the network operation data including at least one of a network transmission characteristic of a network device in the first time period and a first value corresponding to a specified parameter;

[0015] inputting the network operation data into a prediction model;

[0016] Obtaining a second value of the specified parameter output by the prediction model based on the network operation data;

[0017] The designated parameter of the network device in a second time period is configured as the second value.

[0018] In a possible implementation, before inputting the network operation data into the prediction model, the method further includes:

[0019] Obtain historical network operation data corresponding to at least one historical time period, wherein the historical network operation data corresponding to each historical time period in the at least one historical time period includes at least one of the network transmission characteristics of the network device in the historical time period and the value corresponding to the specified parameter, and the at least one historical time period is a time period before the first time period.

[0020] In a possible implementation, inputting the network operation data into a prediction model; and obtaining a second value of the specified parameter output by the prediction model based on the network operation data includes:

[0021] Input historical network operation data corresponding to the at least one historical time period and the network operation data into the prediction model, and obtain a second value of the specified parameter output by the prediction model based on the historical network operation data corresponding to the at least one historical time period and the network operation data, wherein the at least one historical time period is a time period before the first time period.

[0022] In one possible implementation, the method further includes:

[0023] Inputting the network operation data into an initial prediction model, training the initial prediction model, and obtaining the prediction model; or,

[0024] The network operation data is input into the prediction model to obtain an updated prediction model.

[0025] In a possible implementation, before inputting the network operation data into the prediction model, the method further includes:

[0026] Acquire historical network operation data corresponding to a plurality of historical time periods, wherein the historical network operation data corresponding to each of the plurality of historical time periods includes at least one of a network transmission characteristic of the network device in the historical time period and a value corresponding to the specified parameter, and the plurality of historical time periods are time periods before the first time period;

[0027] The historical network operation data corresponding to the multiple historical time periods are input into an initial prediction model, and the initial prediction model is trained to obtain the prediction model.

[0028] In a possible implementation, the network transmission feature includes bandwidth utilization, and before training the initial prediction model to obtain the prediction model, the method further includes:

[0029] Constructing a configuration space including at least one association relationship, wherein each association relationship in the at least one association relationship is used to indicate a value corresponding to a specified parameter;

[0030] Establishing a first feedback relationship and a second feedback relationship, wherein the first feedback relationship is used to indicate the impact of the amount of packets cached by the network device on network transmission performance, and the second feedback relationship is used to indicate the impact of the bandwidth utilization of the network device on network transmission performance;

[0031] The initial prediction model is constructed according to the configuration space, the first reward relationship and the second reward relationship.

[0032] In a possible implementation, the network transmission feature includes at least one of the amount of packets cached by the network device in the first time period, efficiency information of packet transmission, and an average size of transmitted packets.

[0033] In a second aspect, a network parameter configuration method is provided, the method comprising:

[0034] Acquire network operation data corresponding to a first time period, the network operation data including at least one of a network transmission characteristic of a network device in the first time period and a first value corresponding to a specified parameter;

[0035] Inputting the network operation data into a network traffic type identification model;

[0036] Obtaining a first network traffic type of the network device in the first time period output by the network traffic type identification model based on the network operation data, where a network traffic type is used to indicate a message transmission situation in a preset time period;

[0037] Query at least one preset association relationship to obtain a third value of a specified parameter corresponding to the first network traffic type, wherein each association relationship in the at least one association relationship is used to indicate that a network traffic type corresponds to the third value of the specified parameter, and the third value of the specified parameter is a preferred value of the specified parameter under the network traffic indicated by the network traffic type;

[0038] The designated parameter of the network device in a second time period is configured as the third value, where the second time period is a time period after the first time period.

[0039] The method identifies the first network traffic type in the first time period through the network traffic type identification model, and determines the third value of the specified parameter corresponding to the first network traffic type according to at least one association relationship, and configures the third value of the specified parameter corresponding to the first network traffic type as the value of the specified parameter in the second time period. In this configuration method, there is no need to consider the application scenario, and there is no need to manually modify or configure the parameters.

[0040] In a possible implementation, before inputting the network operation data into the network traffic type identification model, the method further includes:

[0041] Establishing the at least one association relationship;

[0042] Acquire at least one historical data set, each of the at least one historical data set includes a plurality of historical network operation data, each of the plurality of historical network operation data includes at least one of a network transmission characteristic of the network device in a historical time period under a network traffic type and a value corresponding to the specified parameter, the historical time period being a time period before the first time period;

[0043] The network traffic type identification model is obtained based on the at least one association relationship, the at least one historical data set and the initial identification model.

[0044] In a possible implementation, the acquiring the network traffic type identification model based on the at least one association relationship, the at least one historical data set, and the initial identification model includes:

[0045] Based on the at least one association relationship, adding a network traffic type tag to each historical network operation data in the at least one historical data set, wherein the network traffic type tag is used to indicate a network traffic type;

[0046] The historical network operation data carrying the network traffic type label is input into the initial recognition model, and the initial recognition model is trained to obtain the network traffic type recognition model.

[0047] In a third aspect, a network parameter configuration device is provided, which is used to execute the network parameter configuration method. Optionally, the network parameter configuration device includes a functional module for executing the network parameter configuration method provided by the first aspect or any optional manner of the first aspect.

[0048] In a fourth aspect, a network parameter configuration device is provided, which is used to execute the network parameter configuration method. Optionally, the network parameter configuration device includes a functional module for executing the network parameter configuration method provided by the second aspect or any optional manner of the second aspect.

[0049] In a fifth aspect, a computer device is provided, which includes a processor and a memory, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement operations performed by the above-mentioned network parameter configuration method.

[0050] In a sixth aspect, a storage medium is provided, in which instructions are stored, and the instructions are loaded and executed by a processor to implement operations performed by the above-mentioned network parameter configuration method.

[0051] In a seventh aspect, a network parameter configuration method is provided, the method being applied to a network parameter configuration system, the network parameter configuration system comprising a network device and a computing device; the method comprising:

[0052] The network device acquires network operation data corresponding to a first time period, where the network operation data includes at least one of a network transmission characteristic of the network device in the first time period and a first value corresponding to a specified parameter;

[0053] The network device sends the network operation data to the computing device;

[0054] The computing device inputs the network operation data into a prediction model; and obtains a second value of the specified parameter output by the prediction model based on the network operation data;

[0055] The computing device sends the second value to the network device;

[0056] The network device configures the designated parameter of the network device in a second time period to the second value, where the second time period is a time period after the first time period.

[0057] In a possible implementation, before the computing device inputs the network operation data into the prediction model, the method further includes:

[0058] The network device acquires historical network operation data corresponding to at least one historical time period, wherein the historical network operation data corresponding to each historical time period in the at least one historical time period includes at least one of a network transmission feature of the network device in the historical time period and a value corresponding to the specified parameter, and the at least one historical time period is a time period before the first time period;

[0059] In a possible implementation, the computing device inputs the network operation data into a prediction model; and obtaining a second value of the specified parameter output by the prediction model based on the network operation data includes:

[0060] The computing device inputs the historical network operation data corresponding to the at least one historical time period and the network operation data into the prediction model, and obtains the second value of the specified parameter output by the prediction model based on the historical network operation data corresponding to the at least one historical time period and the network operation data, wherein the at least one historical time period is a time period before the first time period.

[0061] In a possible implementation, the method further includes:

[0062] The computing device inputs the network operation data into an initial prediction model, trains the initial prediction model, and obtains the prediction model; or,

[0063] The computing device inputs the network operation data into the prediction model to obtain an updated prediction model.

[0064] In a possible implementation, before the computing device inputs the network operation data into the prediction model, the method further includes:

[0065] The network device acquires historical network operation data corresponding to a plurality of historical time periods, the historical network operation data corresponding to each of the plurality of historical time periods comprising at least one of a network transmission feature of the network device in the historical time period and a value corresponding to the specified parameter, the plurality of historical time periods being time periods before the first time period;

[0066] The network device sends the historical network operation data corresponding to the multiple historical time periods to the computing device;

[0067] The computing device inputs the historical network operation data corresponding to the multiple historical time periods into an initial prediction model, trains the initial prediction model, and obtains the prediction model.

[0068] In a possible implementation manner, the network transmission feature includes bandwidth utilization, and before training the initial prediction model to obtain the prediction model, the method further includes:

[0069] The computing device constructs a configuration space including at least one association relationship, each association relationship in the at least one association relationship is used to indicate a value corresponding to a specified parameter;

[0070] The computing device establishes a first feedback relationship and a second feedback relationship, wherein the first feedback relationship is used to indicate the impact of the amount of packets cached by the network device on network transmission performance, and the second feedback relationship is used to indicate the impact of the bandwidth utilization of the network device on network transmission performance;

[0071] The computing device constructs the initial prediction model according to the configuration space, the first reward relationship, and the second reward relationship.

[0072] In a possible implementation, the network transmission feature includes at least one of the amount of packets cached by the network device in the first time period, efficiency information of packet transmission, and an average size of transmitted packets.

[0073] In an eighth aspect, a network parameter configuration method is provided, the method being applied to a network parameter configuration system, the network parameter configuration system comprising a network device and a computing device; the method comprising:

[0074] The network device acquires network operation data corresponding to a first time period, where the network operation data includes at least one of a network transmission characteristic of the network device in the first time period and a first value corresponding to a specified parameter;

[0075] The network device sends the network operation data to the computing device;

[0076] The computing device inputs the network operation data into a network traffic type identification model;

[0077] The computing device obtains the first network traffic type of the network device in the first time period output by the network traffic type identification model based on the network operation data, where a network traffic type is used to indicate the message transmission status in a preset time period;

[0078] The computing device queries at least one preset association relationship to obtain a third value of a specified parameter corresponding to the first network traffic type, wherein each association relationship in the at least one association relationship is used to indicate that a network traffic type corresponds to the third value of the specified parameter, and the third value of the specified parameter is a preferred value of the specified parameter under the network traffic indicated by the network traffic type;

[0079] The computing device sends the second value to the network device;

[0080] The network device configures the designated parameter of the network device in a second time period to the third value, where the second time period is a time period after the first time period.

[0081] In a possible implementation, before the computing device inputs the network operation data into the network traffic type identification model, the method further includes:

[0082] The computing device establishes the at least one association relationship;

[0083] The computing device acquires at least one historical data set, each of the at least one historical data set includes a plurality of historical network operation data, each of the plurality of historical network operation data includes at least one of a network transmission characteristic of the network device in a historical time period under a network traffic type and a value corresponding to the specified parameter, and the historical time period is a time period before the first time period;

[0084] The computing device obtains the network traffic type identification model based on the at least one association relationship, the at least one historical data set, and the initial identification model.

[0085] In a possible implementation, the computing device acquiring the network traffic type identification model based on the at least one association relationship, the at least one historical data set, and the initial identification model includes:

[0086] The computing device adds a network traffic type tag to each historical network operation data in the at least one historical data set based on the at least one association relationship, wherein the network traffic type tag is used to indicate a network traffic type;

[0087] The computing device inputs historical network operation data carrying network traffic type labels into the initial recognition model, trains the initial recognition model, and obtains the network traffic type recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 is a schematic diagram of an ECN mechanism provided in an embodiment of the present application;

[0089] Figure 2 is a schematic diagram of a network parameter configuration system provided in an embodiment of the present application;

[0090] Figure 3 is a schematic diagram of another network parameter configuration system provided in an embodiment of the present application;

[0091] Figure 4 is a schematic diagram of another network parameter configuration system provided in an embodiment of the present application;

[0092] Figure 5 is a structural schematic diagram of a computer device provided in an embodiment of the present application;

[0093] Figure 6 is a flow chart of a network parameter configuration method provided by an embodiment of the present application;

[0094] Figure 7 is a schematic diagram of a negatively correlated piecewise function provided in an embodiment of the present application;

[0095] Figure 8 is a schematic diagram of a network parameter configuration process provided by an embodiment of the present application;

[0096] Fig. 9 is a flow chart of a network parameter configuration method provided by an embodiment of the present application;

[0097] Fig.10is a flowchart of another network parameter configuration method provided in an embodiment of the present application;

[0098] Fig.11 is a flow chart of a network parameter configuration method provided by an embodiment of the present application;

[0099] Fig.12 It is a structural diagram of a network parameter configuration device provided in an embodiment of the present application;

[0100] Fig.13 It is a structural diagram of a network parameter configuration device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0101] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0102] In order to facilitate understanding of the specific implementation process of this application, some terms involved in this application are explained as follows:

[0103] Network parameters: parameters configured in this application. The network parameters of the current time period are the parameters that the network device has configured and is used in the current time period. The network parameters of the next time period are the parameters to be configured by the network device, that is, the parameters to be used in the next time period. The network parameters may include at least one of an ECN parameter, a priority-based flow control (PFC) parameter, and a packet discard probability. Among them, the ECN parameters may include a first threshold, a second threshold, and a marking probability threshold, wherein the first threshold is less than the second threshold. When the network device negotiates the ECN capability with the sender, if the negotiation is successful, the network device may set an ECN mark for its cached messages according to the ECN marking probability. Taking marking messages as an example, see Figure 1 The schematic diagram of an ECN mechanism provided by an embodiment of the present application is shown, wherein: Figure 1 There is a mathematical relationship between the ECN marking probability y and the amount of packets x cached by the network device as shown in formula (1), where n is the first threshold, m is the second threshold, and a is the marking probability threshold.

[0104]

[0105] According to formula (1), the ECN mechanism can be: when the amount of messages cached in the network device is less than or equal to the first threshold n, the network device does not set the ECN mark for the cached messages, and the ECN mark probability can be regarded as 0 at this time; when the amount of messages cached in the network device is greater than the first threshold n and less than or equal to the second threshold m, the network device can determine the ECN mark probability y corresponding to the cached message amount x according to the cached message amount x and the mark probability threshold a, wherein the cached message amount and the determined ECN mark probability are linearly positively correlated, and the determined ECN mark probability is less than or equal to the mark probability threshold, and then the network device can set the ECN mark for the cached messages according to the determined ECN mark probability; when the amount of messages cached in the network device is greater than the second threshold, the network device will set the ECN mark for all cached messages, and the ECN mark probability can be regarded as 1 at this time, that is, 100% mark.

[0106] Network traffic type: used to indicate the message transmission status of the network device within a preset time period. The embodiment of the present application does not specifically limit the preset time period.

[0107] Figure 2 is a schematic diagram of a network parameter configuration system provided in an embodiment of the present application, see Figure 2 The network parameter configuration system 200 may include a sending device 201, a network device 202, and a receiving device 203. The sending device 201 is used to send a message to the network device 202, the network device 202 is used to forward the message sent by the sending device to the receiving device 203, and the receiving device 203 is used to receive the message forwarded by the network device 202.

[0108] The network device 202 is also used to obtain the network transmission characteristics and network parameters of the network device 202 in the current time period in real time, and based on the obtained network transmission characteristics and network parameters, use the prediction model to predict the value of the network parameters of the network device 202 in the next time period, and configure the network parameters of the next time period to the values ​​predicted by the prediction model.

[0109] In one embodiment, the processing process of the network device in the network parameter configuration system is as follows: Figure 3 shown. Figure 3The acquisition module of the network device in the present invention acquires the network transmission characteristics and network parameters of the network device in the current time period, and uploads the acquired network transmission characteristics and network parameters to the parameter calculation module, wherein the parameter calculation module may be a field programmable gate array (FPGA), a processor (central processing units, CPU), a network processor (NP) or an artificial intelligence (AI) chip, and the parameter calculation module uses a prediction model to predict the network parameters of the network device in the next time period based on the network transmission characteristics and network parameters uploaded by the acquisition module, and sends the predicted network parameters, and the network device configures the network parameters of the next time period as the network parameters sent by the parameter calculation module.

[0110] The network device 202 may be a switch or a router, and the embodiment of the present application does not limit the network device 202.

[0111] In some embodiments, the network device 202 is also used to obtain in real time the network transmission characteristics and network parameters of the network device 202 in the current time period, and based on the network traffic identification model and the network transmission characteristics and network parameters of the current time period, obtain the network traffic type of the current time period, determine the value of the network parameter corresponding to the network traffic type according to the obtained network traffic type, and set the network parameter of the next time period to the value corresponding to the network traffic type.

[0112] In some embodiments, the network parameter configuration system 200 may also include a computing device 204, which is used to receive the network transmission characteristics and network parameters of the current time period sent by the network device 202, and based on the received network transmission characteristics and network parameters, use a prediction model to predict the network parameters of the network device 202 in the next time period, and return the predicted network parameters of the network device 202 in the next time period to the network device 202, so that the network device 202 can directly configure the network parameters of the next time period as the network parameters returned by the computing device 204.

[0113] In some embodiments, the computing device 204 may also receive the network transmission characteristics and network parameters of the current time period sent by the network device 202, and obtain the network traffic type of the current time period based on the network traffic identification model and the network transmission characteristics and network parameters of the current time period, determine the network parameters corresponding to the network traffic type according to the obtained network traffic type, and return the determined network parameters to the network device 202, so that the network device 202 can directly configure the network parameters of the next time period as the network parameters returned by the computing device 204. For example, in one embodiment, the processing process of the network device in the network parameter configuration system is as follows: Figure 4 As shown, Figure 4 The collection module of the network device in collects the network transmission characteristics and network parameters of the network device 202 in the current time period, and uploads the collected network transmission characteristics and network parameters to the computing device 204. The computing device 204 uses the prediction model to predict the network parameters of the network device in the next time period based on the network transmission characteristics and network parameters uploaded by the collection module, and sends the predicted network parameters to the network device 202, which configures the network parameters of the next time period as the network parameters sent by the computing device 204.

[0114] The network device 202 and the computing device 204 may be Figure 5 The computer equipment shown, Figure 5 5 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. The computer device 500 may have relatively large differences due to different configurations or performances, and may include one or more CPUs 501 and one or more memories 502, wherein the memory 502 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 501 to implement the methods provided in the following various method embodiments. Of course, the computer device 500 may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The computer device 500 may also include other components for implementing device functions, which will not be described in detail here.

[0115] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions, which can be executed by a processor in a computer device to complete the following network parameter configuration method. For example, the computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0116] In one implementation, the process of configuring network parameters according to the prediction model can be seen in Figure 6 The method described may include the following steps 601-605.

[0117] 601. A network device obtains network operation data corresponding to a first time period, where the network operation data includes at least one of a network transmission characteristic of the network device in the first time period and a first value corresponding to a specified parameter.

[0118] The first time period is the current time period. The first time period may include at least two acquisition moments. The duration of the first time period is the duration from the first acquisition moment to the last acquisition moment of the at least two acquisition moments. For example, if the first time period includes acquisition moments t0 and t1, the duration of the first time period is t1-t0. In the present application, the network device is to configure the network parameters of the time period after the first time period. Therefore, the duration of the first time period can also be considered as a configuration cycle, that is, the network parameter configuration is performed once after the duration of the first time period. Optionally, the duration of different time periods can be set as needed, and the duration of different time periods can be different.

[0119] The designated parameter is also a network parameter, which may include at least one of an ECN parameter, a PFC parameter, and a packet discard probability. The first value corresponding to the designated parameter is also the value of the designated parameter of the network device in the first time period. When the designated parameter is an ECN parameter, the first value of the ECN parameter includes the configuration value of the first threshold, the second threshold, and the marking probability threshold in the first time period. It should be noted that the embodiment of the present application only takes the network parameter as an example for explanation, and the configuration method of other network parameters is the same as the configuration parameter of the ECN parameter. Here, the embodiment of the present application does not repeat the configuration method of other network parameters.

[0120] The network transmission characteristics of the network device in the first time period may be characteristics of message transmission of the network device in the first time period. The network transmission characteristics include the amount of messages cached by the network device in the first time period, efficiency information of message transmission, and the average size of the transmitted messages. The message amount is the number of messages.

[0121] In some embodiments, the network device may selectively obtain individual data in the network operation data, and then combine the individual data into the network operation data. In one possible implementation, this step 601 may be implemented through the process shown in the following steps 11-13.

[0122] Step 11. The network device obtains the amount of messages cached by the network device in the first time period based on the cache parameters of the first collection moment and the last collection moment in the first time period. The cache parameter of the first collection moment is used to indicate the amount of messages cached by the network device at the first collection moment, and the cache parameter of the last collection moment is used to indicate the amount of messages cached by the network device at the last collection moment.

[0123] Optionally, this step 11 can be implemented through the process shown in the following steps 1A-1C.

[0124] Step 1A: The network device performs abnormal value processing on the cache parameters at the first collection time and the cache parameters at the last collection time to obtain the cache parameters at the first collection time and the cache parameters at the last collection time without abnormality.

[0125] Optionally, a Tukey test can be used to determine whether the cache parameter at the first acquisition moment and the cache parameter at the last acquisition moment are abnormal values, or the network device is preset with a normal cache parameter interval, when the cache parameter at the first acquisition moment or the cache parameter at the last acquisition moment is a value within the normal cache parameter interval, the cache parameter at the first acquisition moment or the cache parameter at the last acquisition moment is not an abnormal value, otherwise, it is an abnormal value. The embodiment of the present application does not specifically limit the value range within the normal cache parameter interval.

[0126] When the cache parameter at the first collection moment or the cache parameter at the last collection moment is an abnormal value, the amount of messages cached by the network device in the previous time period of the first time period can be used as the amount of messages cached in the first time period. Alternatively, the network device replaces the abnormal cache parameter (abnormal value) with a preset cache parameter, so that a cache parameter without abnormality can be obtained. The preset cache parameter can be a cache parameter that meets the application scenario, and the embodiment of the present application does not specifically limit the preset cache parameter.

[0127] It should be noted that the network device may also adopt other methods to process outliers for the above cache parameters, and the embodiment of the present application does not specifically limit the method for processing outliers.

[0128] Step 1B: The network device performs exponential sliding average processing on the cache parameters at the first collection time without abnormalities and the cache parameters at the last collection time to obtain the amount of messages cached in the first time period.

[0129] Optionally, an exponential moving average algorithm may be used to average the cache parameters at the first acquisition moment without anomalies and the cache parameters at the last acquisition moment to obtain the first message volume. The exponential moving average algorithm may be expressed using formula (2), where Q is the message volume cached in the first time period, α is the smoothing index, Q(i) is the cache parameter at the first acquisition moment, and Q(j) is the cache parameter at the last acquisition moment.

[0130] Q=α·Q(j)+(1-α)·Q(i) (2)

[0131] Step 1C: normalize the amount of messages cached in the first time period to obtain a normalized message amount value, where the normalized message amount value is used to indicate the amount of messages cached in the first time period.

[0132] In some embodiments, in order to ensure that the packet volume and throughput utilization are in the same dimension in subsequent calculations, the network device may execute step 1C, normalize the packet volume cached in the first time period, and use the normalized value of the packet volume obtained after normalization to replace the packet volume cached in the first time period.

[0133] Step 12: The network device obtains the value of the specified parameter in the first time period, the efficiency information of the message transmission, and the average size of the transmitted messages.

[0134] The network device may determine the value of the specified parameter used in the first time period as the value of the specified parameter in the first time period. The message transmission efficiency information may include the amount of messages with ECN marks set per second, the ECN mark ratio, and the bandwidth utilization.

[0135] The network device may first obtain the amount of its ECN-marked messages in a first time period, and then obtain the amount of ECN-marked messages per second according to the amount of ECN-marked messages in the first time period and the duration of the first time period.

[0136] The network device may first obtain the amount of messages that are ECN marked in the first time period and the amount of messages forwarded in the first time period, and then obtain an ECN marking ratio based on the amount of messages that are ECN marked in the first time period and the amount of messages forwarded in the first time period, where the ECN marking ratio is the ratio of the amount of messages that are ECN marked in the first time period to the amount of messages forwarded in the first time period, for example, θ1=N_e / N_p, where θ1 is the ECN marking ratio, N_e is the amount of messages that are ECN marked in the first time period, and N_p is the amount of messages forwarded in the first time period.

[0137] The network device can first obtain the total number of bytes of messages forwarded in the first time period and the inherent bandwidth ratio of its port, then obtain the number of bytes of messages forwarded per unit time based on the total number of bytes and the length of the first time period, and finally obtain the bandwidth utilization rate based on the number of bytes of messages forwarded per unit time and the inherent bandwidth.

[0138] The network device can first obtain the total number of bytes of messages forwarded in the first time period and the amount of messages forwarded in the first time period, and based on the total number of bytes of messages forwarded in the first time period and the amount of messages forwarded in the first time period, obtain the average size of the transmitted messages, the average size of the transmitted messages being the ratio of the total number of bytes of messages forwarded in the first time period to the amount of messages forwarded in the first time period, for example, θ2=N_b*N_p, where θ2 is the average size of the transmitted messages.

[0139] It should be noted that, when the specified parameters are network parameters other than the ECN parameters, the message transmission efficiency information also includes relevant data on the impact of other network parameters on the message transmission efficiency.

[0140] It should be noted that, in some embodiments, when a network device simultaneously forwards transmission control protocol (TCP) type messages and remote direct memory access (ROCE) type messages through a converged Ethernet, the network device can scale the bandwidth utilization, and the network operation data can also include the scaled bandwidth utilization, wherein the scaled bandwidth utilization is represented by β, β=ε*γ, ε is the bandwidth utilization, and γ is the proportion of ROCE type message traffic configured for the network device.

[0141] Step 13: The network device groups the amount of messages cached in the first time period, the value of the specified parameter in the first time period, the efficiency information of message transmission, and the average size of the transmitted messages into network operation data corresponding to the first time period.

[0142] The network operation data may be in a matrix form, and the data in the matrix is ​​the data acquired by the network device in steps 11-12. The embodiment of the present application does not impose any optional limitation on the combination of the data in the matrix, nor does it impose any limitation on the optional process of combining the data into the network operation data.

[0143] 602. The network device determines a prediction model, where the prediction model is used to predict a value of a specified parameter.

[0144] The prediction model may be a deep reinforcement learning model. The network device may first construct an initial prediction model and then train the initial prediction model to obtain the prediction model. In one possible implementation, step 602 may be implemented through the process shown in steps 21-25 below.

[0145] Step 21: The network device constructs a configuration space including at least one association relationship, where each association relationship in the at least one association relationship is used to indicate a value corresponding to a specified parameter.

[0146] The value corresponding to the specified parameter indicated in at least one association relationship is the value corresponding to various preset specified parameters. Each association relationship may include the serial number of the association relationship and the value of the specified parameter corresponding to the association relationship. Taking the configuration space shown in Table 1 as an example, in Table 1, association relationship 1 includes serial number 1 and the corresponding values ​​of the specified parameter corresponding to serial number 1 are 7, 50 and 1. At this time, the specified parameter is an ECN parameter, 7 is the first threshold, 50 is the second threshold, and 0.01 is the marking probability threshold.

[0147] Table 1

[0148]

[0149] Step 22: The network device establishes a first reporting relationship and a second reporting relationship. The first reporting relationship is used to indicate the impact of the amount of packets cached by the network device on network transmission performance, and the second reporting relationship is used to indicate the impact of the bandwidth utilization of the network device on network transmission performance.

[0150] When the network device cannot forward the received message in time, it will cache the unforwarded message first, and then forward the cached message later. If the network device caches a large number of messages, it means that the network device may have network congestion at this time, that is, the network transmission performance of the network device at this time may be low. Therefore, the first return relationship can be a negative correlation between the first return value and the amount of messages cached by the network device. The negative correlation can be: as the amount of messages cached by the network device increases, the first return value decreases in a step-by-step manner, for example Figure 7 The schematic diagram of a negatively correlated piecewise function provided in the embodiment of the present application is shown in FIG. Figure 7In the example, the maximum value of the first reward value is 1 and the minimum value is 0. As the amount of messages cached by the network device increases, the first reward value decreases in a step-by-step manner within the interval [0,1]. Of course, the negative correlation can also be that as the amount of messages cached by the network device increases linearly, the first reward value decreases linearly. That is, when the amount of messages cached by the network device is larger, the lower the first reward value, and the lower the network transmission performance of the network device will be. Conversely, the higher the first reward value, the higher the network transmission performance of the network device will be. The embodiment of the present application does not limit the optional situation in which the first reward value is negatively correlated with the amount of messages cached by the network device.

[0151] When the bandwidth utilization rate of the network device is higher, it means that the transmission performance of the network device is higher. Conversely, the transmission performance of the network device is lower. Then the second return relationship can be a positive correlation between the second return value and the bandwidth utilization rate. The positive correlation can be: as the bandwidth utilization rate increases, the second return value increases in a step-by-step manner. The positive correlation can also be that as the bandwidth utilization rate increases linearly, the second return value decreases linearly. That is, when the bandwidth utilization rate is higher, the higher the second return value, the higher the transmission performance of the network device. Conversely, the lower the second return value, the lower the transmission performance of the network device. The embodiment of the present application does not limit the optional situation in which the first return value and the bandwidth utilization rate are positively correlated.

[0152] In some embodiments, the network device may also establish a third reporting relationship, which is used to indicate the impact of the service performance indicator of the network device on the network transmission performance. The service performance indicator may be a transmission delay, etc., and the embodiment of the present application does not make optional restrictions on the service performance indicator.

[0153] The closer the current business performance indicator is to the preset indicator, the smaller the difference between the business performance indicator and the preset indicator is, which proves that the current network transmission performance of the network device is better, wherein the business performance indicator can be an indicator of the number of read and write operations per second (input / output operations per second, IOPS), and the preset indicator is an indicator value set in advance. The third feedback relationship can be that the difference between the business performance indicator and the preset indicator is negatively correlated with the third feedback value, that is, the smaller the difference between the business performance indicator and the preset indicator, the larger the third feedback value, and vice versa. The negative correlation is similar to the negative correlation indicated by the first feedback relationship. Here, the embodiment of the present application does not make an optional limitation on the negative correlation.

[0154] Step 23: The network device constructs an initial prediction model according to the configuration space, the first feedback relationship, and the second feedback relationship.

[0155] The network device may first obtain a target feedback relationship based on the first feedback relationship and the second feedback relationship, and the target feedback relationship may be used to indicate the impact of the amount of packets cached by the network device and the bandwidth utilization on the network transmission performance. Then, the network device may construct an initial prediction model based on the target feedback relationship and the configuration space.

[0156] The target feedback relationship may be determined by the relationship between the target feedback value and the first feedback value and the second feedback value. When the first feedback value and / or the second feedback value are larger, the target echo value is larger and the network transmission performance is higher.

[0157] The target reward relationship can be expressed by the following formula (3), where Reward is the target reward value, R thput is the first return value, R qdepth is the second return value, w1 and w2 are weight parameters, and the values ​​of w1 and w2 can be set according to the service preference for throughput latency. The embodiment of the present application does not limit the specific values ​​of w1 and w2.

[0158] Reward=w1*R thput +w2*R qdepth (3)

[0159] The reward relationship can also be expressed by the following formula (4), where w3 is the weight parameter, R kqi is the third return value.

[0160] Reward=w1*R thput +w2*R qdepth +w3*R kqi (4)

[0161] The process of constructing the initial prediction model based on the target reward relationship and the configuration space of the network device may be: based on the target reward relationship and the configuration space, constructing the initial prediction model including an action network and an execution module, wherein the action network is used to assign the value of the specified parameter in the configuration space to each input network environment data, the execution module is used to simulate the message transmission environment, the value of the specified parameter assigned by the action network affects the simulated message transmission environment, and based on the influence of the value of the specified parameter on the message transmission environment, calculate the target reward value and output the network operation data under the affected message transmission environment to the action network. The action network may be a neural network. The embodiment of the present application does not make any optional limitation on the action network.

[0162] Step 24, the network device obtains historical network operation data corresponding to multiple historical time periods, the historical network operation data corresponding to each of the multiple historical time periods includes the network transmission characteristics of the network device in the historical time period and the value corresponding to the specified parameter, and the multiple historical time periods are time periods before the first time period.

[0163] The multiple historical time periods may be time periods when the network device has not used the network parameter configuration method provided in the embodiment of the present application. The historical network operation data corresponding to the multiple historical time periods may be regarded as training samples for model training, and the process of the network device obtaining the historical network operation data corresponding to each historical time period is similar to the process of obtaining the network operation data corresponding to the first time period.

[0164] Step 25: The network device trains the initial prediction model to obtain a prediction model.

[0165] The network device may train the initial prediction model through an offline training process or an online training process. The offline training process may be: the network device inputs historical network operation data corresponding to multiple historical time periods into the initial prediction model, trains the initial prediction model, and obtains the prediction model.

[0166] The process of online training may be: the network device inputs the network operation data corresponding to the first time period into the initial prediction model, trains the initial prediction model, and obtains the prediction model; or the network device inputs the network operation data corresponding to the first time period into the prediction model, obtains the updated prediction model, and the updated prediction model is used to predict the fourth value of the designated parameter of the network device in the third time period based on the corresponding network operation data of the second time period, so that the network device configures the indicated parameter of the third time period to the fourth value. The second time period is the time period after the first time period, and the third time period is the time period after the second time period.

[0167] For the convenience of description, the historical network operation data corresponding to multiple historical time periods and the network operation data corresponding to the first time period are collectively referred to as target network operation data, the target network operation data is input into the initial prediction model, and the initial prediction model is trained to obtain the prediction model. The process can be: initialize the model parameters in the action network; input the target network operation data into the action network, and start the preset number of trainings. Each time the training is performed, the action network assigns a specified parameter value to the input target network operation data, and inputs the maximum value of the assigned specified parameter into the execution module. The execution module uses the received specified parameter value to affect the simulated message transmission environment, and outputs the new network operation data and the target return value under the affected message transmission environment. The action network updates the model parameters of the action network based on the target return value returned by the execution module. When the model parameters are updated, the new network operation data is used as the input of the action network to start the next training. When the preset number of trainings is reached or the target return value output by the execution module reaches the expected value, the training ends. After the above training process, the model parameters of the action network in the initial training model are updated, so the initial training model is also updated to the preset model.

[0168] It should be noted that when the prediction model is obtained based on an offline training process, this step 602 can be executed once before or after step 601, without having to be executed multiple times. When the training model is obtained based on an online training process, the process of obtaining the initial recognition model shown in steps 21-23 only needs to be executed once before or after step 601, without having to be executed multiple times, while step 24 needs to be executed each time a model prediction is performed.

[0169] It should be noted that when the network parameter configuration system includes a computing device, the process shown in step 602 can also be performed by the computing device, wherein the process of obtaining historical network operation data in step 24 can be: the computing device receives the historical network operation data sent by the network device to obtain the historical network operation data.

[0170] 603. The network device inputs the network operation data into the prediction model.

[0171] The network device uses the network operation data of the first time period as input data for the prediction model, and inputs the network operation data of the first time period into the prediction model. The prediction model predicts the value of the specified parameter of the network device in the second time period based on the input network operation data, and outputs the predicted second value of the specified parameter of the second time period.

[0172] It should be noted that, when the prediction model is a model obtained through offline training, the network device can directly execute this step 603. When the prediction model is a model obtained through online training, the network device can input the network operation data into the prediction model to obtain an updated prediction model, that is, the network operation data is regarded as a training sample, and the prediction model is trained to obtain an updated prediction model. Then, the network device no longer executes this step 603, but regards the network operation data as input data of the updated prediction model, inputs the network operation data into the updated prediction model, and the updated prediction model predicts the value of the specified parameter of the network device in the second time period based on the input network operation data, and outputs the second value of the predicted specified parameter in the second time period. When the prediction model is a model obtained through online training, the network device can also, after executing this step 603, regard the network operation data as a training sample, input the network operation data into the prediction model, and obtain an updated prediction model, so that in the second time period, the network device can input the network operation data corresponding to the second time period into the updated prediction model, so that the updated prediction model outputs the predicted fourth value of the specified parameter of the network device in the third time period based on the network operation data corresponding to the second time period.

[0173] 604. The network device obtains a second value of a specified parameter output by the prediction model based on the network operation data.

[0174] When the prediction model outputs a second value of the specified parameter based on the network operation data, the network device can directly obtain the second value.

[0175] The process of predicting the second value of a specified parameter output by a prediction model based on network operation data can be: at least one network operation data is input into an action network in the prediction model, and the action network assigns a value of a specified parameter in the configuration space to each input network operation data based on the current model parameters, and outputs the maximum value of the assigned specified parameter to the prediction model as the second value of the specified parameter.

[0176] It should be noted that the process shown in the above steps 603-604 is a process in which the network device obtains the second value of the specified parameter of the network device in the second time period based on the network operation data corresponding to the first time period and the prediction model. When the network parameter configuration system includes a computing device, the process can also be performed by the computing device. When steps 603-604 are performed by the computing device, after executing step 604, the computing device can send the second value of the specified parameter obtained by the computing device to the network device. When the network device receives the second value sent by the computing device, the specified parameter of the second time period can be configured through the following step 605.

[0177] It should be noted that, in some embodiments, the network device can also obtain historical network operation data corresponding to at least one historical time period, and the historical network operation data corresponding to each historical time period in at least one historical time period includes the network transmission characteristics of the network device in the historical time period and the value corresponding to the specified parameter, and at least one historical time period is a time period before the first time period.

[0178] To further indicate the historical time period at this time, for example, the time between when the network device starts to configure the specified parameters using the method provided in the embodiment of the present application and the last acquisition time in the first time period includes time periods 1-5, where time period 5 is also the first time period, and each historical time period can be any time period in time periods 1-4. The process of the network device obtaining the historical network operation data corresponding to the historical time period is the same as the process of obtaining the network operation data corresponding to the first time period. Here, the embodiment of the present application does not elaborate on the process of the network device obtaining the historical network operation data corresponding to the historical time period.

[0179] After the network device obtains historical network operation data corresponding to at least one historical time period, the network device can input the historical network operation data corresponding to at least one historical time period and the network operation data into the prediction model, and obtain a second value of a specified parameter output by the prediction model based on the historical network operation data corresponding to at least one historical time period and the network operation data, wherein at least one historical time period is a time period before the first time period.

[0180] Since the historical network operation data corresponding to at least one historical time period and the network operation data corresponding to the first time period can constitute a large amount of historical data, and a large amount of historical data can better reflect the message transmission situation in the second time period, therefore, when the prediction model predicts the value of the specified parameter of the network device in the second time period based on a large amount of historical data, the accuracy of the model prediction can be improved.

[0181] The process shown in steps 603-604 may be performed by a parameter calculation module in the network device, for example Figure 8 The schematic diagram of a network parameter configuration process provided by an embodiment of the present application is shown in Figure 8The network device obtains network operation data of a first time period and historical network operation data of a historical time period through data collection and preprocessing, wherein preprocessing refers to the process of obtaining network operation data based on the collected data, and the network device can input the network operation data and the historical network operation data into the parameter calculation module through the forwarding chip, the parameter calculation module inputs the network operation data as output data x, and the historical network operation data as input data y, and inputs the input data x and y into the action network in the prediction model, the input layer, the hidden layer and the activation (Relu) layer in the action network make predictions based on the input network operation data and the historical network operation data, obtain the value of the ECN parameter of the second time period, and output the value of the ECN parameter of the second time period through the output layer in the prediction model, the parameter calculation module sends the value of the ECN parameter of the second time period to the forwarding chip, and the forwarding chip forwards the value of the ECN parameter of the second time period to the configuration module, so that the configuration module configures the ECN parameter of the second time period as the value of the ECN parameter of the forwarding chip.

[0182] 605. The network device configures a specified parameter of the network device in a second time period to a second value.

[0183] After the network device obtains the second value output by the prediction model, the specified parameters of the network device in the second time period can be directly configured to the second value. The embodiment of the present application does not make any optional limitations on the manner and process of configuring the specified parameters of the network device. It is only necessary to ensure that the network device can transmit messages based on the second value of the specified parameter within the second time period.

[0184] The method provided by the embodiment of the present application inputs the network operation data corresponding to the first time period into the prediction model, so that the prediction model predicts the value of the specified parameter of the network device in the second time period based on the input network operation data, and configures the specified parameter of the network device in the second time period to the value predicted by the prediction model. In this configuration method, there is no need to consider the application scenario, and there is no need to manually modify or configure the parameters. Therefore, this configuration method is universal. In addition, when the prediction model predicts the value of the specified parameter of the network device in the second time period based on a large amount of historical data, the accuracy of the model prediction can be improved.

[0185] To further explain the process of network equipment predicting the network parameters to be configured based on the prediction model obtained by offline training, see Fig. 9 The flowchart of a network parameter configuration method provided by an embodiment of the present application is shown in FIG. The flow of the method may optionally include the process shown in steps 901-906.

[0186] 901. The network device collects data in multiple historical time periods, and pre-processes the collected data to obtain historical network operation data corresponding to the multiple historical time periods.

[0187] The network device collects cache parameters, the number of messages with ECN marks, the number of forwarded messages, the total number of bytes of forwarded messages, the inherent bandwidth ratio of its port, and the total number of bytes of forwarded messages at the first sampling moment and the last collection moment in any of the multiple historical time periods, and pre-processes the collected data. The process of obtaining the historical network operation data corresponding to any historical time period is the same as the process of obtaining the network historical operation data in steps 12 and 13. Here, the process of pre-processing the collected data in the embodiment of the present application is not repeated.

[0188] When the network device performs the data quantity process of any historical time period on the multiple historical time periods, the network device can obtain the historical network operation data corresponding to the multiple historical time periods.

[0189] 902. Network devices build configuration space.

[0190] The process shown in step 902 is the same as the process shown in step 21, and step 902 is not described in detail in this embodiment of the present application.

[0191] 903. The network device obtains a target reporting relationship according to the first reporting relationship and the second reporting relationship.

[0192] The process shown in this step 903 is introduced in step 23. Here, this embodiment of the application does not repeat this step 903.

[0193] 904. The network device obtains a prediction model according to the historical network operation data, the configuration space, and the target return relationship corresponding to the multiple historical time periods.

[0194] The network device may first construct an initial prediction model according to the configuration space and the total return relationship, and the optional process is shown in step 23. Then, the network device may train the initial prediction model according to the multiple historical time periods to obtain a prediction model, and the optional process is shown in step 25.

[0195] 905. The network device inputs the network operation data corresponding to the first time period into the prediction model. The prediction model predicts the value of the specified parameter of the network device in the second time period based on the network operation data corresponding to the first time period.

[0196] 906. The prediction model outputs a predicted second value of the specified parameter for a second time period.

[0197] To further explain the process of network devices predicting the network parameters to be configured based on the prediction model obtained through online training, see Fig.10 The flowchart process of another network parameter prediction method provided by the embodiment of the present application is shown. The process of the method can optionally include the process shown in steps 1001-1006.

[0198] 1001. The network device constructs a configuration space.

[0199] The process shown in step 1001 is the same as the process shown in step 21, and step 1001 is not described in detail in this embodiment of the present application.

[0200] 1002. The network device obtains a target reporting relationship according to the first reporting relationship and the second reporting relationship.

[0201] The process shown in step 1002 is introduced in step 23, and the embodiment of the present application will not elaborate on step 1002 here.

[0202] 1003. The network device constructs an initial prediction model according to the configuration space and the target return relationship.

[0203] The process shown in step 1003 is introduced in step 23, and the embodiment of the present application will not elaborate on step 1003 here.

[0204] 1004. The network device collects data in a first time period, and pre-processes the collected data to obtain network operation data corresponding to the first time period.

[0205] The process of obtaining the network operation data corresponding to the first time period in step 1004 is the same as the process of obtaining the historical network operation data of the historical time period in step 601. Here, this embodiment of the present application does not elaborate on this step 1004.

[0206] 1005. The network device inputs the network operation data corresponding to the first time period into the initial prediction model, trains the initial prediction model to obtain an updated prediction model, and inputs the network operation data corresponding to the first time period into the updated prediction model. The updated prediction model predicts the value of the specified parameter of the network device in the second time period based on the network operation data corresponding to the first time period.

[0207] 1006. The updated prediction model outputs a second value of the specified parameter predicted for a second time period.

[0208] The network device can execute step 1004 in the second time period to obtain the network operation data corresponding to the second time period, and train the updated prediction model based on the network operation data corresponding to the second time period to obtain a new prediction model, and input the network operation data corresponding to the second time period into the new prediction model, so that the new prediction model can output the fourth value of the specified parameter of the third time period, and so on. The network device executes the process of steps 1004-1006 in each subsequent time period.

[0209] In some embodiments, the network device can also identify the current network traffic type based on the current network operation data, and configure the network parameters based on the identified network traffic type, see Fig.11 A flowchart of a network parameter configuration method provided in an embodiment of the present application is shown, and the process of the method includes steps 1101-1108.

[0210] 1101. The network device establishes at least one association relationship, each of the at least one association relationship is used to indicate that a network traffic type corresponds to a third value of a specified parameter, and the third value of the specified parameter is a preferred value of the specified parameter under the network traffic indicated by the network traffic type.

[0211] For any business scenario, according to business needs, the business scenario may correspond to at least one network traffic type, wherein a network traffic type is used to indicate the message transmission situation within a preset time period. For example, when three sending devices send messages to a network device within a preset time period, and the network device forwards the messages sent by the three sending devices to two receiving devices, the message transmission situation of the network device within the preset time period is also a network traffic type. For another example, when one sending device sends a message to a network device within a preset time period, and the network device forwards the message sent by the one sending device to three receiving devices, the message transmission situation of the network device within the preset time period is a network traffic type. For another example, when a network device can transmit messages for multiple services within different time periods, if a message transmission for one service is performed within a preset time period, the message transmission situation at this time is a network traffic type, and if a message transmission for another service is performed within a preset time period, the message transmission situation at this time is another network traffic type.

[0212] Based on previous parameter configuration experience, the network device can determine the preferred value of the specified parameter under the network traffic indicated by each network traffic type, and store each network traffic type in association with the preferred value of its corresponding specified parameter, so that each network traffic type has an association relationship with a preferred value of the specified parameter.

[0213] It should be noted that, when the network parameter configuration system includes a computing device, the process shown in step 1101 can also be executed by the computing device.

[0214] 1102. The network device obtains at least one historical data set, each of the at least one historical data set includes multiple historical network operation data, each of the multiple historical network operation data includes at least one item of network transmission characteristics of the network device in a historical time period under a network traffic type and a value corresponding to a specified parameter, and the historical time period is a time period before the first time period.

[0215] Since each association relationship corresponds to a network traffic type, at least one association relationship corresponds to at least one network traffic type. For any network traffic type in the at least one network traffic type, the network device obtains multiple historical network operation data under the any network traffic type, and combines the obtained historical network operation data into a historical data set corresponding to the any network traffic type. When the network device performs the above process of obtaining a historical data set for the at least one network traffic type, the network device can finally obtain at least one historical data set.

[0216] The process of the network device acquiring multiple historical network operation data under any network traffic type is similar to the process shown in step 24. The difference from step 24 is that the network device must acquire historical network operation data under any network traffic type, and the network traffic type is not limited in step 24.

[0217] It should be noted that when the network parameter configuration system includes a computing device, after the network device completes step 1102, at least one acquired historical data set can be sent to the computing device, and the computing device executes the process shown in the following steps 1103-1107.

[0218] 1103. The network device obtains a network traffic type identification model based on at least one association relationship, at least one historical data set, and an initial identification model.

[0219] The network device can first determine the network traffic type in the association relationship corresponding to each historical data set, so that for any historical data set, multiple historical network operation data in any historical data set can fully reflect the message transmission situation indicated by the corresponding network traffic type. Then, the network device can use each historical network operation data in each historical data set as a sample, train the initial recognition model, and obtain a network traffic type recognition model for identifying network traffic types. In a possible implementation, this step 1103 can be implemented by the process shown in the following steps 31-32.

[0220] Step 31: The network device adds a network traffic type tag to each historical network operation data in at least one historical data set based on at least one association relationship, where a network traffic type tag is used to indicate a network traffic type.

[0221] Since the historical network operation data in each historical data set are obtained under the network traffic indicated by the network traffic type in an association relationship, for any historical data set, the network device can determine the network traffic type corresponding to the any historical data set, so that the network device can add a label indicating the network traffic type corresponding to the any historical data set in each historical network operation data of the any historical data set, and the label of the network traffic type can be a type identifier of the network traffic type, and different network traffic types have different type identifiers, and each type identifier is used to indicate a network traffic type. The embodiment of the present application does not specifically limit the type identifier and the method of adding labels to the historical network operation data.

[0222] Step 32: The network device inputs the historical network operation data carrying the network traffic type label into the initial recognition model, trains the initial recognition model, and obtains the network traffic type recognition model.

[0223] The initial recognition model may be a neural network model, and the process shown in step 23 may be: the network device initializes the model parameters in the initial recognition model, and inputs the historical network operation data carrying the network traffic type label into the initial recognition model, and the initial recognition model performs a preset number of model training based on the network transmission characteristics, model labels and supervised learning algorithms in each historical network operation data. For each training, the initial recognition model performs type recognition based on the current model parameters and the network transmission characteristics in each historical network operation data, obtains the network traffic type of each historical network operation data, determines whether the network traffic type of each historical network operation data identified is correct according to the network traffic type label carried in each historical network operation data, and counts the accuracy of this recognition result, and then updates the current model parameters of the initial recognition model according to the supervised learning algorithm for the next training, and ends the training when the preset number of training times or the accuracy of the recognition result reaches a preset value. Since the model parameters of the initial recognition model are updated after each training, when the training is completed, the initial recognition model will be updated to the network traffic type recognition model, so that the network traffic model can quickly and accurately recognize the network traffic type of the network operation data based on the input network operation data.

[0224] It should be noted that the supervised learning algorithm can be a random forest algorithm or a neural network supervised learning algorithm. The embodiment of the present application does not specifically limit the supervised learning algorithm. In addition, the process shown in the above steps 1101-1104 only needs to be executed once and does not need to be executed multiple times.

[0225] 1104. The network device obtains network operation data corresponding to the first time period, where the network operation data includes at least one of a network transmission characteristic of the network device in the first time period and a first value corresponding to a specified parameter.

[0226] The process shown in step 1104 is the same as the process shown in step 401, and the process shown in step 1104 is not described in detail in this embodiment of the present application.

[0227] 1105. The network device inputs the network operation data into the network traffic type identification model.

[0228] 1106. The network device obtains a first network traffic type of the network device in a first time period based on a network traffic type identification model outputted from the network operation data. A network traffic type is used to indicate a message transmission situation in a preset time period.

[0229] When the network device inputs the network traffic data into the network traffic type identification model, the network traffic identification model identifies the network traffic type corresponding to the network transmission characteristics based on the current model parameters and the network transmission characteristics included in the network traffic data, that is, the first network traffic type in the first time period, and outputs the identified first network traffic type.

[0230] 1107. The network device queries at least one preset association relationship to obtain a third value of a specified parameter corresponding to the first network traffic type.

[0231] Since at least one association relationship is stored in the network device, the network device can determine the third value of the specified parameter associated with the first network traffic type by querying the at least one stored association relationship.

[0232] It should be noted that when the process shown in step 1107 is executed by a computing device, if step 1107 is completed, the computing device can send the third value of the specified parameter corresponding to the determined first network traffic type to the network device so that the network device executes the following step 1108.

[0233] 1108. The network device configures a network parameter of the network device in a second time period to a third value, where the second time period is a time period after the first time period.

[0234] Since each network traffic type is used to indicate the message transmission status within a preset time period, the first time period and the second time period can be regarded as preset time periods, so that the network device can configure the specified parameters in the second time period as the third value of the specified parameters corresponding to the first network traffic type.

[0235] The method provided in the embodiment of the present application identifies the first network traffic type in the first time period through the network traffic type identification model, and determines the third value of the specified parameter corresponding to the first network traffic type according to at least one association relationship, and configures the third value of the specified parameter corresponding to the first network traffic type as the specified parameter of the second time period. In this configuration method, there is no need to consider the application scenario, and there is no need to manually modify or configure the parameters. Therefore, this configuration method is universal. In addition, since each association relationship is used to indicate that a network traffic type corresponds to the third value of the specified parameter, and the third value of the specified parameter is the preferred value of the specified parameter under the network traffic indicated by the network traffic type, the specified parameter configured in the second time period is also the preferred value of the specified parameter under the network traffic indicated by the first network traffic type, so that the network device transmits messages in the second time period based on the preferred value of the specified parameter, which can improve the transmission efficiency of the message to avoid transmission congestion.

[0236] Fig.12 : is a schematic diagram of the structure of a network parameter configuration device provided in an embodiment of the present application, the device comprising:

[0237] The acquisition unit 1201 is used to execute the above step 601;

[0238] Input unit 1202, used to execute the above step 603;

[0239] The acquisition unit 1201 is also used to execute the above step 604;

[0240] The configuration unit 1203 is used to execute the above step 605.

[0241] Optionally, the acquisition unit 1201 is further configured to execute the above step 24.

[0242] Optionally, the acquisition unit 1201 is used to input historical network operation data corresponding to the at least one historical time period and the network operation data into the prediction model, and obtain the second value of the specified parameter output by the prediction model based on the historical network operation data corresponding to the at least one historical time period and the network operation data, wherein the at least one historical time period is a time period before the first time period.

[0243] Optionally, the device further comprises:

[0244] A first training unit is used to input the network operation data into an initial prediction model, train the initial prediction model, and obtain the prediction model; or,

[0245] An updating unit is used to input the network operation data into the prediction model to obtain an updated prediction model.

[0246] Optionally, the device further comprises a second training unit;

[0247] The acquisition unit 1201 is also used to execute the above step 24;

[0248] The second training unit is used to input the historical network operation data corresponding to the multiple historical time periods into the initial prediction model, train the initial prediction model, and obtain the prediction model.

[0249] Optionally, the device further comprises:

[0250] A construction unit for executing the above step 21;

[0251] Establishing a unit for executing the above step 22;

[0252] The construction unit is also used to execute the above step 23.

[0253] Optionally, the network transmission characteristics include at least one of the amount of packets cached by the network device in the first time period, efficiency information of packet transmission, and an average size of transmitted packets.

[0254] Fig.13 : is a schematic diagram of the structure of a network parameter configuration device provided in an embodiment of the present application, the device comprising:

[0255] An acquisition unit 1301 is used to execute the above step 1104;

[0256] An input unit 1302, used to execute the above step 1105;

[0257] The acquisition unit 1301 is used to execute the above step 1106;

[0258] A query unit 1303, configured to execute the above step 1103;

[0259] The configuration unit 1304 is used to execute the above step 1104.

[0260] Optionally, the device further comprises:

[0261] Establishing a unit for executing the above step 1101;

[0262] The acquisition unit 1301 is also used to execute the above step 1102;

[0263] The acquisition unit 1301 is also used to execute the above step 1103.

[0264] Optionally, the acquisition unit 1301 is used to execute the above steps 31-32.

[0265] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.

[0266] It should be noted that: the network parameter configuration device provided in the above embodiment only uses the division of the above functional modules as an example when performing the network parameter configuration. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the network parameter configuration method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0267] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0268] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A network parameter configuration method, characterized in that: The method is applied to a network device, the network device includes a forwarding chip and a parameter calculation module, and the method includes: The parameter calculation module obtains the network transmission characteristics of the network device, where the network transmission characteristics are characteristics of message transmission of the network device; The parameter calculation module inputs the network transmission characteristics into a prediction model; The parameter calculation module outputs a configuration value of a specified parameter using the prediction model based on the network transmission characteristics, the specified parameter including an explicit congestion notification (ECN) parameter, and the specified parameter is used to set an ECN mark for a message forwarded by the network device; The parameter calculation module sends the configuration value to the forwarding chip; The forwarding chip receives the configuration value sent by the parameter calculation module; The specified parameters of the forwarding chip are configured as the configuration values.

2. The method according to claim 1, characterized in that The parameter calculation module includes one of a field programmable gate array, a central processing unit, a network processor or an artificial intelligence chip.

3. The method according to claim 1 or 2, characterized in that: The parameter calculation module obtains the network transmission characteristics of the network device, including: The parameter calculation module receives the network transmission characteristics of the network device sent by the forwarding chip, wherein the network transmission characteristics are obtained by preprocessing the acquired collected data by the forwarding chip.

4. The method according to claim 1 or 2, characterized in that: The network transmission feature includes at least one of bandwidth utilization, cache message volume, or cache message volume normalization value.

5. The method according to claim 1 or 2, characterized in that: The ECN parameters include a first threshold, a second threshold, and a marking probability threshold, wherein the first threshold is smaller than the second threshold.

6. The method according to claim 1 or 2, characterized in that: The prediction model is a deep reinforcement learning model.

7. A network device, characterized in that: It includes a forwarding chip and a parameter calculation module, wherein: The forwarding chip is used to perform the operation performed by the forwarding chip of the network device in the method according to any one of claims 1 to 6; The parameter calculation module is used to execute the operations performed by the parameter calculation module of the network device in the method according to any one of claims 1 to 6.

8. A communication system, comprising a network device, wherein the network device is used to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are executed, a computer implements the method according to any one of claims 1 to 6.

Images