Congestion control method, device and equipment for intelligent calculation center network and storage medium

Abstract

The invention relates to a congestion control method and device for an intelligent calculation center network, equipment and a storage medium. The method comprises the following steps: acquiring a current message sending rate of a QP queue corresponding to an RP terminal under the condition of determining that the RP terminal receives a CNP message in a first target period; determining a speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in the historical period; wherein the historical period is a period before the first target period; determining a target message sending rate according to the speed reduction parameter and the current message sending rate; and controlling the RP terminal to send the QP message in the QP queue according to the target message sending rate in the first target period. By adopting the method, the speed reduction parameter can be determined in real time according to the queue length of the QP queue in the historical period, so that the message sending rate of the RP terminal is dynamically adjusted, different adjustment amplitudes are adopted in different periods, the method is suitable for network scenes with different congestion degrees, and the message sending rate is more accurately controlled.

Description

Technical Field

[0001] The present application relates to the field of intelligent computing network technology, and in particular to a congestion control method, device, equipment and storage medium for an intelligent computing center network. Background Art

[0002] With the rapid development of intelligent computing, the traffic of intelligent computing centers is becoming increasingly complex, especially the massive data throughput in large model training, which brings more urgent traffic control and congestion control problems to intelligent computing centers. At present, mainstream intelligent computing centers generally use RoCE (RDMA over Converged Ethernet, RDMA technology based on Ethernet) v2 protocol to build RDMA (Remote Direct Memory Access) network, and RoCE v2 protocol uses DCQCN (Data Center Quantized Congestion Notification) algorithm for congestion control.

[0003] In the DCQCN algorithm, the switch marks the data packets in the QP (Queue Pair) queue with ECN (Explicit Congestion Notification) with a certain probability, so that the receiving end of the data packet will mark the data packet sent to the source sending end as CNP (Congestion Notification Packet) packet. In this way, the source sending end can perceive that this terminal is the main source or one of the main sources of congestion in the network of the intelligent computing center, and then quickly reduce its own message sending rate.

[0004] However, the current method of reducing the message sending rate on the terminal side is relatively simple and cannot adapt to network scenarios with different congestion levels, resulting in poor congestion control effects. Summary of the invention

[0005] Based on this, it is necessary to provide a congestion control method, device, equipment and storage medium for the intelligent computing center network that can adapt to network scenarios with different congestion levels and accurately control the message sending rate in response to the above technical problems.

[0006] In a first aspect, the present application provides a congestion control method for an intelligent computing center network, comprising:

[0007] When it is determined that the RP terminal receives the CNP message within the first target period, the current message sending rate of the QP queue corresponding to the RP terminal is obtained; wherein the RP terminal is any terminal in the intelligent computing center network;

[0008] Determine the speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in the historical period; wherein the historical period is the period before the first target period;

[0009] Determine the target message sending rate according to the speed reduction parameter and the current message sending rate;

[0010] The RP terminal is controlled to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0011] In one embodiment, the historical cycle includes a first historical cycle and a second historical cycle, the first historical cycle is a cycle before the first target cycle, and the second historical cycle is a cycle before the first historical cycle;

[0012] According to the queue length of the QP queue in the historical period, the speed reduction parameter corresponding to the first target period is determined, including:

[0013] The speed reduction parameter corresponding to the first target period is determined according to the size relationship between the queue length of the QP queue in the first historical period and the queue length of the QP queue in the second historical period.

[0014] In one embodiment, determining the speed reduction parameter corresponding to the first target period according to the relationship between the queue length of the QP queue in the first historical period and the queue length of the QP queue in the second historical period includes:

[0015] If the queue length of the QP queue in the first historical period is less than the queue length of the QP queue in the second historical period, the set value is used as the speed reduction parameter corresponding to the first target period;

[0016] If the queue length of the QP queue in the first historical period is not less than the queue length of the QP queue in the second historical period, the speed reduction parameter corresponding to the first target period is determined according to the ratio between the queue length of the QP queue in the first historical period and the queue length of the QP queue in the second historical period.

[0017] In one embodiment, determining a target message sending rate according to the speed reduction parameter and the current message sending rate includes:

[0018] Determine the nonlinear parameter according to the first power; wherein the first power is a power with e as the base and the speed reduction parameter as the exponent;

[0019] The ratio between the current message sending rate and the nonlinear parameter is determined as the target message sending rate.

[0020] In one of the embodiments, after controlling the RP terminal to send the QP message in the QP queue according to the target message sending rate within the first target period, the method further includes:

[0021] When it is determined that the RP terminal does not receive the CNP message within the second target period, and the second target period is a speed-up period, determine the target period number corresponding to the second target period; wherein the second target period is the target period number period after the first target period during which the RP terminal continues to fail to receive the CNP message; the speed-up period is the period during which the RP terminal restores the message sending rate to the current message sending rate;

[0022] The target message sending rate is updated according to the target cycle number and the preset speed-up cycle number;

[0023] The RP terminal is controlled to send the QP messages in the QP queue according to the updated target message sending rate within the second target period.

[0024] In one embodiment, updating the target message sending rate according to the target cycle number and the preset speed-up cycle number includes:

[0025] Determine the speed-up parameter corresponding to the second target period according to the difference between the target period number and the preset speed-up period number;

[0026] Determine the adaptive parameter according to the second power; wherein the second power is a power with e as the base and the speed increase parameter as the exponent;

[0027] The product of the adaptive parameter and the target message sending rate is determined as the updated target message sending rate.

[0028] In a second aspect, the present application also provides a congestion control device for an intelligent computing center network, including:

[0029] The current rate acquisition module is used to obtain the current message sending rate of the QP queue corresponding to the RP terminal when it is determined that the RP terminal receives the CNP message within the first target period; wherein the RP terminal is any terminal in the intelligent computing center network;

[0030] A speed reduction parameter determination module, used to determine the speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in a historical period; wherein the historical period is a period before the first target period;

[0031] A target rate determination module, used to determine a target message sending rate according to a speed reduction parameter and a current message sending rate;

[0032] The message sending control module is used to control the RP terminal to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0033] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0034] When it is determined that the RP terminal receives the CNP message within the first target period, the current message sending rate of the QP queue corresponding to the RP terminal is obtained; wherein the RP terminal is any terminal in the intelligent computing center network;

[0035] Determine the speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in the historical period; wherein the historical period is the period before the first target period;

[0036] Determine the target message sending rate according to the speed reduction parameter and the current message sending rate;

[0037] The RP terminal is controlled to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0038] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the following steps are implemented:

[0039] When it is determined that the RP terminal receives the CNP message within the first target period, the current message sending rate of the QP queue corresponding to the RP terminal is obtained; wherein the RP terminal is any terminal in the intelligent computing center network;

[0040] Determine the speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in the historical period; wherein the historical period is the period before the first target period;

[0041] Determine the target message sending rate according to the speed reduction parameter and the current message sending rate;

[0042] The RP terminal is controlled to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0043] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0044] When it is determined that the RP terminal receives the CNP message within the first target period, the current message sending rate of the QP queue corresponding to the RP terminal is obtained; wherein the RP terminal is any terminal in the intelligent computing center network;

[0045] Determine the speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in the historical period; wherein the historical period is the period before the first target period;

[0046] Determine the target message sending rate according to the speed reduction parameter and the current message sending rate;

[0047] The RP terminal is controlled to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0048] The congestion control method, device, equipment and storage medium of the above-mentioned intelligent computing center network, when determining that the RP terminal receives the CNP message within the first target period, obtains the current message sending rate of the QP queue corresponding to the RP terminal; wherein the RP terminal is any terminal in the intelligent computing center network; according to the queue length of the QP queue in the historical period, determines the speed reduction parameter corresponding to the first target period; wherein the historical period is the period before the first target period; according to the speed reduction parameter and the current message sending rate, determines the target message sending rate; controls the RP terminal to send the QP message in the QP queue at the target message sending rate within the first target period. The present application determines the speed reduction parameter in real time through the queue length of the QP queue in the historical period, and then dynamically adjusts the message sending rate of the RP terminal based on the speed reduction parameter when it is detected that the RP terminal is one of the terminals causing network congestion, so as to realize different adjustment ranges in different periods. Compared with the prior art, the present application can adapt to network scenarios with different congestion levels and control the message sending rate more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0050] FIG1 is an application environment diagram of a congestion control method for an intelligent computing center network in one embodiment;

[0051] FIG2 is an architecture diagram of an intelligent computing terminal device for implementing a congestion control method for an intelligent computing center network in one embodiment;

[0052] FIG3 is a flow chart of a congestion control method for an intelligent computing center network in one embodiment;

[0053] FIG4 is a schematic diagram of a process for determining a speed reduction parameter in one embodiment;

[0054] FIG5 is a schematic diagram of a process for determining a target message sending rate in one embodiment;

[0055] FIG6 is a schematic diagram of a process for increasing the message sending rate in one embodiment;

[0056] FIG7 is a schematic diagram of a process for updating a target message sending rate in one embodiment;

[0057] FIG8 is a flow chart of a congestion control method for an intelligent computing center network in another embodiment;

[0058] FIG9 is a structural block diagram of a congestion control device of an intelligent computing center network in one embodiment;

[0059] FIG10 is a diagram showing the internal structure of a computer device in one embodiment;

[0060] FIG. 11 is a diagram showing the internal structure of a computer device in another embodiment. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0062] The congestion control method for the intelligent computing center network provided in the embodiment of the present application can be applied to the application environment shown in Figure 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Both the terminal 102 and the server 104 can independently or interactively execute the congestion control method for the intelligent computing center network, and obtain the current message sending rate of the QP queue corresponding to the RP (Reaction Point) terminal when determining that the RP terminal receives the CNP message within the first target period; wherein the RP terminal is any terminal in the intelligent computing center network; according to the queue length of the QP queue in the historical period, determine the speed reduction parameter corresponding to the first target period; wherein the historical period is the period before the first target period; according to the speed reduction parameter and the current message sending rate, determine the target message sending rate; control the RP terminal to send the QP message in the QP queue at the target message sending rate within the first target period.

[0063] The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. Head-mounted devices may be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server 104 may be an independent physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0064] In an exemplary embodiment, as shown in FIG2 , an intelligent computing terminal device is provided, which can be configured on a terminal or a server to execute the steps of the congestion control method of the intelligent computing center network in the present application.

[0065] Specifically, the intelligent computing terminal device includes an intelligent computing center terminal RP algorithm core processing unit, a QP queue buffer area, a timer, a parameter register, a speed increase / deceleration calculation engine, a QP queue sending rate controller and a sending adapter, etc.

[0066] Among them, the core processing unit of the RP algorithm of the intelligent computing center terminal is used to process congestion perception and the functions of the RP terminal side of the DCQCN algorithm, and coordinate and control the logical business process operation of other units; the QP queue buffer area is the buffer area where the terminal network card stores QP messages, including the QP sending queue and the QP receiving queue. The core processing unit of the RP algorithm of the intelligent computing center terminal can monitor the QP receiving queue and sense whether the RP terminal has received the CNP message; the timer is a configurable timing device for controlling the sending and congestion perception related operations in the QP queue buffer area; the parameter register is used to store a series of key parameters of the DCQCN algorithm on the RP terminal side, such as the timer duration parameter, etc. The parameters stored in the parameter register are The number can be modified by the RP algorithm core processing unit of the intelligent computing center terminal, and is usually valid in one timing cycle; the speed increase / deceleration calculation engine is used to increase or decrease the message sending rate of the specified QP queue under the control of the RP algorithm core processing unit of the intelligent computing center terminal, and can read relevant parameters from the parameter register; the QP queue sending rate controller controls the rate of the specified QP sending queue of the sending adapter based on the calculated message sending rate, that is, the target message sending rate; the sending adapter is connected to the sending port of the intelligent computing terminal as a QP message sending channel, and the sending adapter controls the rate of the specified QP sending queue under the management of the QP queue sending rate controller.

[0067] In an exemplary embodiment, as shown in FIG3 , a congestion control method for an intelligent computing center network is provided, and the method is described by taking the application of the method to the intelligent computing terminal device in FIG2 as an example, including:

[0068] S301 : When it is determined that the RP terminal receives a CNP message within a first target period, a current message sending rate of a QP queue corresponding to the RP terminal is obtained.

[0069] Among them, there are three roles in the DCQCN algorithm, namely, RP terminal, which represents the sender network card, CP (Congestion Point) terminal, which represents the switch, and NP (Notification Point) terminal, which represents the receiver network card. This embodiment implements the adjustment of the message sending rate by executing the corresponding algorithm on the RP terminal side. The RP terminal is any terminal in the intelligent computing center network. The CNP message is a data message sent to the RP terminal by the receiving end that receives the data message carrying the ECN congestion mark to notify the traffic speed reduction. The QP queue is a virtual interface between hardware and software, which stores the tasks sent by the software to the hardware in sequence. The QP queue corresponding to the RP terminal includes a QP receiving queue and a QP sending queue, which respectively store the data messages received by the RP terminal and the data messages to be sent.

[0070] Specifically, the work of receiving / sending data messages of the RP terminal is periodically sensed through a timer, and the parameter register can store the duration parameter of the timer. In each cycle, it is detected whether the RP terminal receives a CNP message. In this way, if it is detected that the RP terminal receives a CNP message in the current cycle, it is determined that the RP terminal is one of the terminals that cause congestion in the network of the intelligent computing center, and the message sending rate of the RP terminal needs to be reduced. Then, the current cycle is used as the first target cycle, and the message sending rate of the QP queue corresponding to the RP terminal in the first target cycle is recorded, that is, the current message sending rate.

[0071] The current message sending rate may be the message sending rate of the QP queue recorded at the start time of the first target period, or the message sending rate of the QP queue recorded at the time when the RP terminal receives the CNP message, or the average message sending rate from the start time of the first target period to the time when the RP terminal receives the CNP message, which is not limited in this embodiment. The method of recording the current message sending rate in each period remains consistent.

[0072] S302: Determine a speed reduction parameter corresponding to a first target period according to a queue length of a QP queue in a historical period.

[0073] Among them, the historical period is the period before the first target period, and the historical period can be one period or multiple periods. Generally, the closer the historical period is to the first target period, the higher the accuracy of the speed reduction parameter determined according to the queue length of the QP queue in the historical period, and thus the better the effect of reducing the message sending rate.

[0074] Exemplarily, when the historical period is the period before the first target period, there is a positive nonlinear relationship between the queue length of the QP queue in the historical period and the speed reduction parameter corresponding to the first target period, that is, the longer the queue length of the QP queue in the historical period, the larger the speed reduction parameter corresponding to the first target period, and the growth rate of the speed reduction parameter increases with the growth of the queue length.

[0075] Exemplarily, when the historical period is the first three periods of the first target period, the speed reduction parameter corresponding to the first target period can be determined according to the statistical values ​​of the QP queue lengths in these three periods, such as the sum, average, weighted average, standard deviation or variance of the queue lengths, so as to realize data analysis of the queue length and its change of the QP queue in the historical period, so that the longer the queue length of the QP queue in the historical period, the larger the speed reduction parameter corresponding to the first target period, and the growth rate of the speed reduction parameter changes with the change of the statistical value.

[0076] It should be noted that, for each cycle, the queue length of the QP queue in the cycle may refer to the queue length at the start time of the cycle, the queue length at the end time of the cycle, or the average queue length in the cycle, and this embodiment does not limit this. The method for determining the queue length of the QP queue in each cycle remains consistent.

[0077] S303: Determine a target message sending rate according to the speed reduction parameter and the current message sending rate.

[0078] The speed reduction parameter can be used to directly or indirectly measure the nonlinear relationship between the current message sending rate and the target message sending rate. The target message sending rate can be directly calculated based on the speed reduction parameter and the current message sending rate.

[0079] Exemplarily, an exponential function is formed based on the speed reduction parameter, and the current message sending rate is substituted into the exponential function to calculate the target message sending rate. The exponential function can avoid a collapse-type speed reduction and avoid bandwidth waste.

[0080] S304: Control the RP terminal to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0081] The QP message in the QP queue sent by the RP terminal is the data message in the QP sending queue corresponding to the RP terminal.

[0082] After determining the target message sending rate corresponding to the first target period, the RP terminal is controlled to continue sending QP messages according to the target message sending rate until the first target period ends. If the timer reaches the corresponding timing duration, the next period is entered.

[0083] Optionally, if the RP terminal still receives CNP messages in the next cycle, the next cycle is used as the new first target cycle, and the steps of executing the congestion control method of the intelligent computing center network are returned until the RP terminal no longer receives CNP messages. Generally, after the RP terminal no longer receives CNP messages, the message sending rate of the RP terminal needs to be increased to restore it to the level before the speed reduction process.

[0084] In the congestion control method of the above-mentioned intelligent computing center network, when it is determined that the RP terminal receives the CNP message within the first target period, the current message sending rate of the QP queue corresponding to the RP terminal is obtained; according to the queue length of the QP queue in the historical period, the speed reduction parameter corresponding to the first target period is determined; wherein the historical period is the period before the first target period; according to the speed reduction parameter and the current message sending rate, the target message sending rate is determined; and the RP terminal is controlled to send the QP message in the QP queue according to the target message sending rate within the first target period. This embodiment determines the speed reduction parameter in real time by the queue length of the QP queue in the historical period, and then when it is detected that the RP terminal is one of the terminals causing network congestion, the message sending rate of the RP terminal is dynamically reduced based on the speed reduction parameter, so as to realize different speed reduction amplitudes in different periods. Compared with the linear speed reduction method in the prior art, this application can adapt to network scenarios with different congestion levels and control the message sending rate more accurately.

[0085] In an exemplary embodiment, the historical cycle includes a first historical cycle and a second historical cycle, the first historical cycle is a cycle before the first target cycle, and the second historical cycle is a cycle before the first historical cycle. The above S302 includes: determining the speed reduction parameter corresponding to the first target cycle according to the relationship between the queue length of the QP queue in the first historical cycle and the queue length of the QP queue in the second historical cycle.

[0086] Specifically, if the queue length of the QP queue in the first historical period is smaller than the queue length of the QP queue in the second historical period, it means that the network congestion caused by the RP terminal is gradually decreasing, and it is likely that the queue length of the QP queue in the first target period is also smaller than the queue length of the QP queue in the first historical period. A smaller speed reduction parameter can be determined to make the speed reduction trend in the first target period relatively gentle.

[0087] If the queue length of the QP queue in the first historical period is greater than the queue length of the QP queue in the second historical period, it means that the network congestion caused by the RP terminal is gradually increasing. It is likely that the queue length of the QP queue in the first target period is also greater than the queue length of the QP queue in the first historical period. A larger speed reduction parameter can be determined, so that the speed reduction trend in the first target period is steeper.

[0088] If the queue length of the QP queue in the first historical period is equal to the queue length of the QP queue in the second historical period, it is likely that the queue length of the QP queue in the first target period is also relatively close, which means that the effect of reducing the message sending rate of the RP terminal is not obvious enough, and a relatively large reduction parameter can be determined.

[0089] In this embodiment, by measuring the size relationship between the queue lengths of the QP queues in the first two cycles of the first target cycle, the congestion level caused by the increase in queue transmission in each cycle can be adaptively evaluated, so as to more accurately determine the speed reduction parameters corresponding to the first target cycle.

[0090] In an exemplary embodiment, as shown in FIG4 , the above-mentioned determining the speed reduction parameter corresponding to the first target period according to the size relationship between the queue length of the QP queue in the first historical period and the queue length of the QP queue in the second historical period includes:

[0091] S401, determining whether the queue length of the QP queue in the first historical period is less than the queue length of the QP queue in the second historical period.

[0092] If yes, execute S402; if no, execute S403.

[0093] S402: Use the set value as a deceleration parameter corresponding to the first target period.

[0094] The set value is a constant, which can be, but is not limited to, 1.

[0095] S403 : Determine a speed reduction parameter corresponding to the first target period according to a ratio between a queue length of the QP queue in the first historical period and a queue length of the QP queue in the second historical period.

[0096] Exemplarily, Q1 represents the queue length of the QP queue in the first historical period, and Q2 represents the queue length of the QP queue in the second historical period. When QI / Q2 is less than 1, the speed reduction parameter corresponding to the first target period is determined to be 1. When QI / Q2 is not less than 1, a suitable parameter is introduced, and the speed reduction parameter corresponding to the first target period is calculated in combination with QI / Q2, such as presetting a constant α, and taking the product of QI / Q2 and α as the speed reduction parameter corresponding to the first target period.

[0097] As an optional implementation in this embodiment, the deceleration parameter corresponding to the first target period is calculated using the following formula:

[0098] ;

[0099] Among them, γ is the speed reduction parameter, Q(t-1) is the queue length of the QP queue in the first historical period, Q(t-2) is the queue length of the QP queue in the second historical period, and α is a constant.

[0100] In this embodiment, by comparing the queue length of the QP queue in the first historical period with the queue length of the QP queue in the second historical period, the speed reduction parameter in the first target period is determined according to different situations. When the queue length of the QP queue gradually decreases, a quantitative speed reduction parameter is used to determine the target message sending rate. When the queue length of the QP queue gradually increases, the greater the increase, the greater the speed reduction parameter, which greatly reduces the message sending rate, thereby alleviating network congestion more quickly.

[0101] In an exemplary embodiment, as shown in FIG5 , the above S303 includes:

[0102] S501, determining a nonlinear parameter according to a first power.

[0103] The first power is a power with e as the base and the speed reduction parameter as the exponent. The speed reduction parameter is represented by γ. The first power can be represented by e γ .

[0104] According to the first power, a related nonlinear parameter is determined to characterize the ratio between the current message sending rate and the target message sending rate.

[0105] Exemplarily, a constant is added to the first power to obtain the corresponding nonlinear parameter, which is expressed as 1+e γ .

[0106] S502: Determine the ratio between the current message sending rate and the nonlinear parameter as the target message sending rate.

[0107] For example, the current message sending rate is represented by R T , the nonlinear parameter is expressed as 1+e γ , the target message sending rate is represented by R C , we can get R C =R T / 1+e γ .

[0108] In this embodiment, a nonlinear parameter is further determined based on the speed reduction parameter, and the target message sending rate corresponding to the first target period is obtained in combination with the current message sending rate. This can improve the accuracy and rationality of the target message sending rate and optimize the step of speed reduction processing for the message sending rate of the RP terminal.

[0109] In an exemplary embodiment, as shown in FIG6 , after the above S304, the congestion control method of the above intelligent computing center network further includes:

[0110] S601: When it is determined that the RP terminal does not receive a CNP message within a second target period and the second target period is a speed-up period, determine a target period number corresponding to the second target period.

[0111] The second target period is the target period number of periods after the first target period during which the RP terminal continues to fail to receive the CNP message, and the speed-up period is the period during which the RP terminal restores the message sending rate to the current message sending rate.

[0112] Exemplarily, the RP terminal continuously receives CNP messages from the first cycle to the third cycle, and the corresponding target message sending rate is repeatedly determined in these three cycles, and the message sending rate of the RP terminal is reduced until the RP terminal does not receive the CNP message in the fourth cycle, and it is determined that the network congestion caused by the RP terminal is alleviated at this time, and the speed reduction process can be stopped, and the speed increase process stage is entered. That is, the fourth cycle is the first cycle in which the RP terminal continuously does not receive the CNP message, the fourth cycle is the speed increase cycle, and the fourth cycle can be used as the second target cycle, and the corresponding target cycle number is 1.

[0113] Furthermore, if the RP terminal still does not receive the CNP message within the 5th cycle, and the message sending rate of the RP terminal has not recovered to the level before the speed reduction processing, then the 5th cycle is the second cycle in which the RP terminal has not received the CNP message. The 5th cycle is the speed increase cycle. The 5th cycle can be used as the second target cycle, and the corresponding target cycle number is 2. The target cycle numbers are calculated in sequence until the message sending rate of the RP terminal recovers to the level before the speed reduction processing.

[0114] S602: Update the target message sending rate according to the target cycle number and the preset speed-up cycle number.

[0115] The preset number of speed-up cycles is a constant, which can be expressed as T.

[0116] It should be noted that if the message sending rate is increased too quickly as soon as the stage of increasing the speed of the message sending rate of the RP terminal is entered, it is easy to cause congestion recovery at the congestion point. Therefore, the DCQCN algorithm adopts a staged speed increase method, that is, different parameters are set for different stages. However, this embodiment provides a new method for determining the target message sending rate, by introducing the target cycle number and the speed increase cycle number, to design the growth inflection point of the message sending rate in the speed increase stage, without setting different parameters for different stages.

[0117] Specifically, according to experiments, the network of the intelligent computing center can generally return to normal after T cycles, and different intelligent computing centers may have different corresponding acceleration cycle numbers T. Before the target cycle number reaches the acceleration cycle number, the message sending rate of the RP terminal is slowly increased to avoid network congestion again. After the target cycle number reaches the acceleration cycle number, the message sending rate of the RP terminal is quickly increased to restore the network status of the intelligent computing center as soon as possible.

[0118] Exemplarily, the speed-up parameter corresponding to the second target period is determined according to the size relationship between the target period number and the speed-up period number, and then the target message sending rate is updated according to the speed-up parameter. The target message sending rate is the target message sending rate determined in the period before the second target period.

[0119] S603: Control the RP terminal to send the QP messages in the QP queue according to the updated target message sending rate within the second target period.

[0120] After determining the updated target message sending rate, the RP terminal is controlled to continue sending QP messages according to the updated target message sending rate until the second target period ends. If the timer reaches the corresponding timing duration, the next period is entered.

[0121] Optionally, if the RP terminal still does not receive the CNP message within the next cycle and the next cycle is still a speed-up cycle, the next cycle is used as a new second target cycle, and the steps of this embodiment are returned to be executed until the message sending rate of the RP terminal is restored to the level before the speed reduction.

[0122] In this embodiment, when it is detected that the RP terminal no longer causes network congestion and the message sending rate of the RP terminal has not been fully recovered, the target cycle number and the preset cycle number can be used to determine the increase in the message sending rate of the RP terminal within the second target cycle, thereby dynamically adjusting the message sending rate of the RP terminal to achieve different adjustment ranges for different cycles. Compared with the prior art, this embodiment can adaptively increase the speed without setting multiple parameters, which is conducive to simplifying the control process of message sending.

[0123] In an exemplary embodiment, as shown in FIG. 7 , the above S602 includes:

[0124] S701, determining a speed-up parameter corresponding to a second target period according to a difference between a target period number and a preset speed-up period number.

[0125] Exemplarily, the target cycle number is represented by t, the speed-up cycle number is represented by T, a constant β is provided, and the product of tT and the constant β is used as the speed-up parameter corresponding to the second target cycle, represented by β*(tT).

[0126] S702: Determine an adaptive parameter according to the second power.

[0127] The second power is a power with e as the base and the speed parameter as the exponent. The second power can be expressed as e β*（t-T） .

[0128] According to the second power, a related adaptive parameter is determined to characterize the ratio between the updated target message sending rate and the target message sending rate before the update.

[0129] Exemplarily, a constant is added to the second power to obtain the corresponding adaptive parameter, which is expressed as 1+e β*（t-T） .

[0130] S703: The product of the adaptive parameter and the target message sending rate is determined as the updated target message sending rate.

[0131] For example, the target message sending rate is represented by R C , the adaptive parameter is expressed as 1+e β*（t-T） , the updated target message sending rate can be expressed as R C =R C (1+e β*（t-T） ).

[0132] In this embodiment, the adaptive parameter is further determined based on the speed-increasing parameter, and the target message sending rate is updated to obtain the updated target message sending rate corresponding to the second target period. This can improve the accuracy and rationality of the updated target message sending rate and optimize the step of speeding up the message sending rate of the RP terminal.

[0133] In an exemplary embodiment, as shown in FIG8 , a congestion control method for an intelligent computing center network is provided, including:

[0134] S801: When it is determined that the RP terminal receives a CNP message within a first target period, a current message sending rate of a QP queue corresponding to the RP terminal is obtained.

[0135] S802: Determine whether the queue length of the QP queue in the first historical period is less than the queue length of the QP queue in the second historical period.

[0136] The first historical cycle is the cycle before the first target cycle, and the second historical cycle is the cycle before the first historical cycle.

[0137] If yes, execute S803; if no, execute S804.

[0138] S803: Use the set value as a speed reduction parameter corresponding to the first target period.

[0139] S804: Determine a speed reduction parameter corresponding to the first target period according to a ratio between a queue length of the QP queue in the first historical period and a queue length of the QP queue in the second historical period.

[0140] S805, determining a first power according to the speed reduction parameter, and determining a nonlinear parameter according to the first power.

[0141] The first power is a power with e as the base and the speed reduction parameter as the exponent.

[0142] S806: Determine the ratio between the current message sending rate and the nonlinear parameter as the target message sending rate.

[0143] S807: Control the RP terminal to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0144] S808: When it is determined that the RP terminal does not receive the CNP message within the second target period and the second target period is a speed-up period, determine a target period number corresponding to the second target period.

[0145] The second target period is the target period number of periods after the first target period during which the RP terminal continues to fail to receive the CNP message; the speed-up period is the period during which the RP terminal restores the message sending rate to the current message sending rate.

[0146] S809: Determine a speed-up parameter corresponding to a second target period according to a difference between the target period number and a preset speed-up period number.

[0147] S810, determining a second power according to the speed-up parameter, and determining an adaptive parameter according to the second power.

[0148] The second power is a power with e as the base and the growth rate parameter as the exponent.

[0149] S811: The product of the adaptive parameter and the target message sending rate is determined as the updated target message sending rate.

[0150] S812: Control the RP terminal to send the QP messages in the QP queue according to the updated target message sending rate within the second target period.

[0151] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0152] Based on the same inventive concept, the embodiment of the present application also provides a congestion control device for an intelligent computing center network for implementing the congestion control method for the intelligent computing center network involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of the congestion control device for one or more intelligent computing center networks provided below can refer to the limitations of the congestion control method for the intelligent computing center network above, and will not be repeated here.

[0153] In an exemplary embodiment, as shown in FIG9 , a congestion control device for an intelligent computing center network is provided, including:

[0154] The current rate acquisition module 10 is used to acquire the current message sending rate of the QP queue corresponding to the RP terminal when it is determined that the RP terminal receives the CNP message within the first target period;

[0155] The speed reduction parameter determination module 20 is used to determine the speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in the historical period; wherein the historical period is the period before the first target period;

[0156] The target rate determination module 30 is used to determine the target message sending rate according to the speed reduction parameter and the current message sending rate;

[0157] The message sending control module 40 is used to control the RP terminal to send the QP messages in the QP queue according to the target message sending rate within the first target period.

[0158] In an exemplary embodiment, the historical cycle includes a first historical cycle and a second historical cycle, the first historical cycle is the previous cycle of the first target cycle, and the second historical cycle is the previous cycle of the first historical cycle; the above-mentioned speed reduction parameter determination module 20 is specifically used to determine the speed reduction parameter corresponding to the first target cycle based on the relationship between the queue length of the QP queue in the first historical cycle and the queue length of the QP queue in the second historical cycle.

[0159] In an exemplary embodiment, the above-mentioned speed reduction parameter determination module 20 includes:

[0160] A first parameter determination unit, configured to use a set value as a speed reduction parameter corresponding to a first target period when a queue length of a QP queue in a first historical period is less than a queue length of a QP queue in a second historical period;

[0161] The second parameter determination unit is used to determine the speed reduction parameter corresponding to the first target period according to the ratio between the queue length of the QP queue in the first historical period and the queue length of the QP queue in the second historical period when the queue length of the QP queue in the first historical period is not less than the queue length of the QP queue in the second historical period.

[0162] In an exemplary embodiment, the target rate determination module 30 includes:

[0163] A third parameter determination unit is used to determine the nonlinear parameter according to the first power number; wherein the first power number is a power number with e as the base and the speed reduction parameter as the exponent;

[0164] The target rate determination unit is used to determine the ratio between the current message sending rate and the nonlinear parameter as the target message sending rate.

[0165] In an exemplary embodiment, the congestion control device of the intelligent computing center network further includes:

[0166] The cycle number determination module is used to determine the target cycle number corresponding to the second target cycle when it is determined that the RP terminal has not received the CNP message within the second target cycle and the second target cycle is a speed-up cycle; wherein the second target cycle is the target cycle number cycle after the first target cycle in which the RP terminal has not continuously received the CNP message; the speed-up cycle is the cycle in which the RP terminal restores the message sending rate to the current message sending rate;

[0167] A rate update module, used to update the target message sending rate according to the target cycle number and the preset speed-up cycle number;

[0168] Correspondingly, the message sending control module 40 is further used to control the RP terminal to send the QP messages in the QP queue according to the updated target message sending rate within the second target period.

[0169] In an exemplary embodiment, the rate updating module includes:

[0170] a fourth parameter determination unit, configured to determine a speed-up parameter corresponding to a second target period according to a difference between the target period number and a preset speed-up period number;

[0171] a fifth parameter determination unit, configured to determine the adaptive parameter according to a second power; wherein the second power is a power with e as the base and a speed increase parameter as the exponent;

[0172] The rate updating unit is used to determine the product of the adaptive parameter and the target message sending rate as the updated target message sending rate.

[0173] Each module in the congestion control device of the above-mentioned intelligent computing center network can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0174] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG10. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as the number of speed-up cycles. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a congestion control method for an intelligent computing center network is implemented.

[0175] In an exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG11. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected via a system bus, and the communication interface, the display unit, and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a congestion control method for an intelligent computing center network is implemented. The display unit of the computer device is used to form a visually visible picture, which may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0176] Those skilled in the art will understand that the structures shown in FIGS. 10 and 11 are merely block diagrams of partial structures related to the scheme of the present application, and do not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than those shown in the figures, or combine certain components, or have a different arrangement of components.

[0177] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of the above-mentioned congestion control method for the intelligent computing center network when executing the computer program.

[0178] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the congestion control method of the above-mentioned intelligent computing center network are implemented.

[0179] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned congestion control method for the intelligent computing center network.

[0180] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0181] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0182] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A congestion control method for an intelligent computing center network, characterized in that: The method comprises: When it is determined that the reaction point RP terminal receives the congestion notification data packet CNP message within the first target period, the current message sending rate of the queue corresponding to the RP terminal to the QP queue is obtained; wherein the RP terminal is any terminal in the intelligent computing center network; Determine, according to the queue length of the QP queue in a historical period, a speed reduction parameter corresponding to the first target period; wherein the historical period is a period before the first target period; Determining a target message sending rate according to the speed reduction parameter and the current message sending rate; The RP terminal is controlled to send the QP messages in the QP queue according to the target message sending rate within the first target period.

2. The method according to claim 1, characterized in that The historical cycle includes a first historical cycle and a second historical cycle, the first historical cycle is a cycle before the first target cycle, and the second historical cycle is a cycle before the first historical cycle; The determining, according to the queue length of the QP queue in the historical period, the speed reduction parameter corresponding to the first target period includes: The speed reduction parameter corresponding to the first target period is determined according to the size relationship between the queue length of the QP queue in the first historical period and the queue length of the QP queue in the second historical period.

3. The method according to claim 2, characterized in that The determining, according to a relationship between a queue length of the QP queue in the first historical period and a queue length of the QP queue in the second historical period, a speed reduction parameter corresponding to the first target period includes: If the queue length of the QP queue in the first historical period is less than the queue length of the QP queue in the second historical period, the set value is used as the speed reduction parameter corresponding to the first target period; If the queue length of the QP queue in the first historical period is not less than the queue length of the QP queue in the second historical period, the speed reduction parameter corresponding to the first target period is determined according to the ratio between the queue length of the QP queue in the first historical period and the queue length of the QP queue in the second historical period.

4. The method according to any one of claims 1 to 3, characterized in that The determining a target message sending rate according to the speed reduction parameter and the current message sending rate includes: Determine the nonlinear parameter according to a first power; wherein the first power is a power with e as the base and the speed reduction parameter as the exponent; The ratio between the current message sending rate and the nonlinear parameter is determined as the target message sending rate.

5. The method according to claim 1, characterized in that After controlling the RP terminal to send the QP messages in the QP queue according to the target message sending rate within the first target period, the method further includes: When it is determined that the RP terminal does not receive the CNP message within the second target period, and the second target period is a speed-up period, determine the target period number corresponding to the second target period; wherein the second target period is the target period number period after the first target period during which the RP terminal continues to fail to receive the CNP message; the speed-up period is the period during which the RP terminal restores the message sending rate to the current message sending rate; Updating the target message sending rate according to the target number of cycles and a preset number of increasing speed cycles; The RP terminal is controlled to send the QP messages in the QP queue according to the updated target message sending rate within the second target period.

6. The method according to claim 5, characterized in that The updating of the target message sending rate according to the target cycle number and the preset speed-up cycle number includes: Determining a speed-up parameter corresponding to the second target period according to a difference between the target period number and a preset speed-up period number; Determine an adaptive parameter according to a second power; wherein the second power is a power with e as the base and the speed increase parameter as the exponent; The product of the adaptive parameter and the target message sending rate is determined as the updated target message sending rate.

7. A congestion control device for an intelligent computing center network, characterized in that: The device comprises: A current rate acquisition module, used to obtain the current message sending rate of the QP queue corresponding to the RP terminal when it is determined that the RP terminal receives the CNP message within the first target period; wherein the RP terminal is any terminal in the intelligent computing center network; A speed reduction parameter determination module, configured to determine the speed reduction parameter corresponding to the first target period according to the queue length of the QP queue in a historical period; wherein the historical period is a period before the first target period; A target rate determination module, configured to determine a target message sending rate according to the speed reduction parameter and the current message sending rate; The message sending control module is used to control the RP terminal to send the QP messages in the QP queue according to the target message sending rate within the first target period.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images