Intelligent calculation center congestion control method and device, computer equipment, medium and product

Abstract

The invention relates to an intelligent calculation center congestion control method and device, computer equipment, a medium and a product. The method comprises the following steps: acquiring the average queue length of a current queue corresponding to a current port, acquiring a speed reduction coefficient corresponding to the current queue according to the average queue length under the condition that the average queue length meets a congestion condition, constructing a direct return congestion notification message based on the speed reduction coefficient and a queue message cached in the current queue, and sending the direct return congestion notification message to the current port. And sending the direct return congestion notification message to a sending end, so as to indicate the sending end to analyze the direct return congestion notification message, obtain the current queue and the speed reduction coefficient, and also indicate the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient. By adopting the method, network congestion can be effectively solved.

Description

Technical Field

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

[0002] In intelligent computing centers, high-throughput scenarios like large-scale model training lead to complex network traffic. In particular, sudden many-to-one data interactions can easily cause network congestion. Achieving efficient congestion control while maintaining high throughput and low latency, while balancing bandwidth utilization and network stability, has become a core challenge that intelligent computing centers urgently need to address.

[0003] Currently, Ethernet-based remote direct access protocols are mainly used to build remote direct access memory networks, and end-to-end congestion control algorithms are used for congestion control. In this process, the switch monitors the port queue length. If it exceeds the preset threshold, a congestion notification mark is added to the IP packet header of the queue pair message. After the receiving end detects the congestion notification mark, it generates a congestion notification packet and returns it to the sending end. The sending end then autonomously adjusts the sending rate of the queue pair.

[0004] However, since the congestion notification mark needs to be generated by the receiving end in the form of a congestion notification packet and then returned to the sending end, the congestion information transmission path is long, and the returned packet itself may be blocked in the congested link, resulting in insufficient speed reduction. Summary of the Invention

[0005] Based on this, it is necessary to provide an intelligent computing center congestion control method, device, computer equipment, medium and product that can effectively solve network congestion 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, comprising:

[0007] Get the average queue length of the current queue corresponding to the current port;

[0008] When the average queue length meets the congestion condition, the speed reduction coefficient corresponding to the current queue is obtained based on the average queue length;

[0009] A direct return congestion notification message is constructed based on the speed reduction coefficient and the queue messages buffered in the current queue. The integer part and the fractional part of the speed reduction coefficient are stored in two reserved fields of the direct return congestion notification message respectively.

[0010] The direct return congestion notification message is sent to the sending end to instruct the sending end to parse the direct return congestion notification message, obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

[0011] In one embodiment, the step of obtaining the average queue length of the current queue corresponding to the current port includes:

[0012] Get the queue length of the current queue corresponding to the current port in the current cycle and the previous average queue length of the current queue in the cycle before the current cycle.

[0013] According to the queue length, the previous average queue length and the first preset value, the average queue length corresponding to the current queue in the current period is obtained.

[0014] In one embodiment, the step of obtaining a speed reduction coefficient corresponding to the current queue based on the average queue length includes:

[0015] Obtain the compression target value; the compression target value is used to represent the target value corresponding to the queue length when the congestion condition is not met;

[0016] Obtain the speed reduction coefficient corresponding to the current queue based on the average queue length, the compression target value, and the second preset value.

[0017] In one embodiment, the step of constructing a direct return congestion notification message based on the speed reduction coefficient and queue messages buffered in the current queue includes:

[0018] Get a copy of the queue message;

[0019] Send the queue message to the receiving end according to the IP address information in the queue message;

[0020] Based on the speed reduction coefficient, the copied message is updated to obtain a direct return congestion notification message.

[0021] In one embodiment, the copied message includes an IP header, a protocol data header, a basic transport header, and a data field; and the step of updating the copied message based on the speed reduction coefficient to obtain a direct return congestion notification message includes:

[0022] Perform address swapping on IP packet headers and port swapping on protocol packet headers;

[0023] The operation type field in the basic transport header is set to a preset value, and the integer part and the fractional part of the speed reduction coefficient are stored in two reserved fields in the basic transport header respectively; the preset value is used to identify the message identity of the direct return congestion notification message;

[0024] Delete the data field.

[0025] In one embodiment, the method further comprises:

[0026] In the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient, when the timer is in a timeout state, obtaining the average queue length of the current queue corresponding to the next cycle of the current cycle;

[0027] If the average queue length corresponding to the next period meets the congestion condition, the process returns to the step of obtaining the speed reduction coefficient corresponding to the current queue based on the average queue length, and continues the process until the average queue length of the current queue no longer meets the congestion condition.

[0028] In a second aspect, the present application further provides an intelligent computing center congestion control device, comprising:

[0029] The queue acquisition module is used to obtain the average queue length of the current queue corresponding to the current port;

[0030] A speed reduction acquisition module is used to obtain a speed reduction coefficient corresponding to the current queue according to the average queue length when the average queue length meets the congestion condition;

[0031] a message construction module, configured to construct a direct return congestion notification message based on a speed reduction coefficient and queue messages buffered in a current queue; wherein an integer part and a fractional part of the speed reduction coefficient are respectively stored in two reserved fields of the direct return congestion notification message;

[0032] The congestion notification module is used to send a direct return congestion notification message to the sending end to instruct the sending end to parse the direct return congestion notification message, obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

[0033] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements any one of the method steps in the first aspect when executing the computer program.

[0034] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements any one of the method steps in the first aspect when the computer program is executed by a processor.

[0035] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which implements any one of the method steps in the first aspect when executed by a processor.

[0036] The above-mentioned intelligent computing center congestion control method, device, computer equipment, medium and product obtain the average queue length of the current queue corresponding to the current port. When the average queue length meets the congestion condition, the speed reduction coefficient corresponding to the current queue is obtained according to the average queue length. Based on the speed reduction coefficient and the queue messages cached in the current queue, a direct return congestion notification message is constructed, and the direct return congestion notification message is sent to the sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient. It also instructs the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient. It can efficiently and accurately transmit the speed reduction instruction and effectively solve network congestion. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. 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 any creative work.

[0038] Figure 1 This is a diagram of an application environment of a congestion control method for an intelligent computing center in one embodiment;

[0039] Figure 2 1. A flow chart of a congestion control method for an intelligent computing center according to an embodiment;

[0040] Figure 3 This is a structural block diagram of an intelligent computing center congestion control system in one embodiment;

[0041] Figure 4 Schematic diagram of a flow chart of a congestion control method for an intelligent computing center in another embodiment;

[0042] Figure 5 This is a structural block diagram of a congestion control device for an intelligent computing center in one embodiment;

[0043] Figure 6 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0045] The congestion control method for the intelligent computing center provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. The sending end 102 communicates with the receiving end 106 via the intelligent computing center switch 104. The intelligent computing center switch 104 is used to obtain the average queue length of the current queue corresponding to the current port. When the average queue length meets the congestion condition, the intelligent computing center switch 104 obtains the speed reduction coefficient corresponding to the current queue based on the average queue length. Based on the speed reduction coefficient and the queue messages cached in the current queue, a direct return congestion notification message is constructed and sent to the sending end 102 to instruct the sending end 102 to parse the direct return congestion notification message, obtain the current queue and the speed reduction coefficient, and also instruct the sending end 102 to adjust the sending rate of the current queue based on the speed reduction coefficient. The sending end 102 and the receiving end 106 can be, but are not limited to, various personal computers, laptops, smart phones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projectors, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head mounted device may be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The intelligent computing center switch 104 may be a network device used in an intelligent computing center.

[0046] It should be noted that the beneficial effects or technical problems solved by the embodiments of the present application are not limited to this one, but may also include other implicit or related problems. For details, please refer to the description of the following embodiments.

[0047] Before introducing the specific embodiments of the present invention, the professional terms involved in the present invention are explained:

[0048] RDMA: Remote Direct Memory Access, remote direct memory access network, allows computers to directly access remote host memory through the network;

[0049] RoCEv2: RDMA over Converged Ethernet, the second generation of Ethernet-based remote direct memory access protocol;

[0050] DCQCN: Data Center Quantized Congestion Notification, a rate-based end-to-end congestion control algorithm;

[0051] Incast: many-to-one communication mode;

[0052] CNP: Congestion Notification Packet, congestion notification packet;

[0053] IB: InfiniBand network;

[0054] ECN: Explicit Congestion Notification, explicit congestion notification;

[0055] PFC: Priority Flow Control, priority flow control;

[0056] SQ: Send Queue, send queue;

[0057] RQ: Receive Queue, receive queue;

[0058] QP: Queue Pair, a queue pair consisting of a sending queue and a receiving queue;

[0059] RP: Reaction Point, reaction point;

[0060] CP: Congestion Point, congestion point;

[0061] UDP: User Datagram Protocol, User Datagram Protocol;

[0062] BTH: Base Transport Header, basic transport header;

[0063] Op Code: operation code.

[0064] With the widespread development of intelligent computing, congestion control has become a key area of focus for data centers. Intelligent computing centers face the challenges of traffic control and congestion control. They have evolved from the north-south traffic of the previous cloud computing era to a direction that prioritizes east-west traffic while also taking into account north-south traffic. This is especially true for the massive data throughput of large model training, and the traffic in intelligent computing centers is becoming increasingly complex. Currently, intelligent computing centers primarily use the RoCEv2 protocol to build their RDMA networks. RoCEv2 carries RDMA based on Ethernet frames and IP protocols, significantly reducing the construction and operation and maintenance costs of IB-based networks and achieving excellent cost-performance. RoCEv2 typically uses the ECN and PFC mechanisms to control congestion in the intelligent computing center network. To ensure network losslessness, the ECN mechanism must be operating effectively before the PFC mechanism takes effect.

[0065] Because switches in intelligent computing centers often experience congestion due to incast traffic, they must monitor the outbound port's buffer queue status in real time to proactively detect and predict the possibility of congestion. Currently, when predicting congestion, these switches employ Random Early Detection (RED) or improved algorithms such as WRED and EA-RED. When the buffer queue exceeds a preset lower limit, these algorithms apply an ECN mark to the IP header of each RDMA QP packet in the queue based on a calculated probability. When an ECN-marked RDMA QP packet reaches the receiving terminal, the receiving terminal detects the ECN mark and constructs a CNP packet within the QP packet returning to the sender, which is then returned to the source. If the source detects that the received packet is a CNP packet, it activates the DCQCN data RP-side algorithm, reducing the sending rate by the corresponding QP queue. Once the congestion improves and no CNP packets are received, the sending rate is gradually increased according to the RP-side algorithm.

[0066] However, after the switch adds the ECN mark to the QP message, the message must continue to be sent to the receiving end, which generates a CNP message. This will cause a delay in the congestion message reaching the source sender. Even the returned CNP message may be congested on the way back. In addition, the source sender starts the RP-side algorithm of DCQCN to reduce and increase the sending rate of QP messages. There is a lack of coordination mechanism with the switch. It autonomously reduces and increases the speed in stages based on multiple parameters, and cannot perceive the congestion situation at the congestion point in real time.

[0067] Based on this, an embodiment of the present application provides a congestion control method for an intelligent computing center. By constructing a direct return CNP message packet and coordinating the source sending end to reduce the sending rate of the QP queue, the end-to-end network collaborative traffic congestion control within the intelligent computing center is realized based on the existing RoCEv2 protocol, especially the network congestion caused by Incast traffic, which can effectively achieve pressure reduction.

[0068] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0069] In an exemplary embodiment, Figure 2 As shown, a congestion control method for an intelligent computing center is provided. Figure 1 Taking the intelligent computing center switch 104 in FIG as an example, the method includes the following steps 202 to 208. Among them:

[0070] Step S202: Obtain the average queue length of the current queue corresponding to the current port.

[0071] Optionally, the current queue of the current port refers to the specific output port on the Intelligent Computing Center switch currently processing traffic and the specific traffic queue managed by that port. The average queue length refers to the average number of packets (queue messages) waiting to be sent in that queue over a specific time period and is a key indicator of queue congestion. The Intelligent Computing Center switch continuously monitors the status of its output ports and on-port queues, calculating the average number of packets in a specific queue over a period of time. This provides a more stable reflection of congestion trends and avoids misjudging transient traffic bursts.

[0072] Step S204: When the average queue length meets the congestion condition, a speed reduction coefficient corresponding to the current queue is obtained according to the average queue length.

[0073] Optionally, the congestion condition refers to the average queue length exceeding a preset threshold, indicating that the queue is accumulating packets and may be at risk of congestion or is already congested. The rate reduction factor specifies the percentage by which the sender should reduce the send rate for that particular queue. Typically, the rate reduction factor is greater than 1, thereby reducing the pressure on the current queue.

[0074] Step S206: construct a direct return congestion notification message based on the speed reduction coefficient and the queue messages buffered in the current queue; the integer part and the fractional part of the speed reduction coefficient are respectively stored in two reserved fields of the direct return congestion notification message.

[0075] Optionally, a direct return congestion notification message is sent directly back by the intelligent computing center switch to the data sender to accurately notify congestion information and speed reduction requirements. Among them, the reserved bits in the direct return congestion notification message that have not yet been defined by the protocol standard can be used for customized purposes. The direct return congestion notification message is constructed by combining the calculated speed reduction coefficient with the message information in the current queue. The speed reduction coefficient in the message is encoded into the message and split into an integer part and a decimal part, which are respectively stored in the originally unused reserved fields in the message protocol header.

[0076] Step S208: Send the direct return congestion notification message to the sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

[0077] Optionally, the intelligent computing center switch sends the constructed message back to the data source (the sender). Upon receiving the message, the sender parses it and, upon identifying it as a congestion notification message, extracts the affected queue ID (i.e., the current queue). It then extracts the integer and decimal parts from the message's two reserved fields, recombining them to obtain the original rate reduction factor. Based on the parsed rate reduction factor, the sender reduces the traffic rate assigned to the target queue.

[0078] In the above-mentioned intelligent computing center congestion control method, by obtaining the average queue length of the current queue corresponding to the current port, when the average queue length meets the congestion condition, the speed reduction coefficient corresponding to the current queue is obtained according to the average queue length, and based on the speed reduction coefficient and the queue messages cached in the current queue, a direct return congestion notification message is constructed, and the direct return congestion notification message is sent to the sending end to instruct the sending end to parse the direct return congestion notification message, obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient. It can efficiently and accurately transmit the speed reduction instruction and effectively solve network congestion.

[0079] In an exemplary embodiment, the step of obtaining the average queue length of the current queue corresponding to the current port includes: obtaining the queue length of the current queue corresponding to the current port in the current period, and the previous average queue length corresponding to the current queue in the period before the current period; and obtaining the average queue length corresponding to the current queue in the current period based on the queue length, the previous average queue length, and a first preset value.

[0080] Optionally, during the current cycle, the number of packets waiting to be sent in the current port's buffer queue is sampled in real time. For example, at the end of the cycle, the instantaneous queue length of the current queue is obtained. Additionally, the previous average queue length for the previous cycle is read from historical records. This average queue length refers to the average value calculated during the previous cycle and represents a smoothed representation of historical congestion trends.

[0081] For example, the average queue length is calculated as: Qavg(t) = (1-w) * Qavg(t-1) + w*Q(t). Qavg(t) is the current average queue length, w is a constant coefficient, Qavg(t-1) is the average queue length of the previous period, and Q(t) is the queue length of the current period. When Qavg(t) exceeds the congestion threshold, congestion risk is present.

[0082] In this embodiment, by obtaining the average queue length corresponding to the current queue in the current cycle based on the previous average queue length in the previous cycle, the current congestion risk can be accurately identified, the probability of false triggering of speed reduction can be reduced, and the accuracy of congestion control can be improved.

[0083] In an exemplary embodiment, the step of obtaining a speed reduction coefficient corresponding to the current queue based on the average queue length includes: obtaining a compression target value; the compression target value is used to represent the target value corresponding to the queue length when the congestion condition is not met; and obtaining the speed reduction coefficient corresponding to the current queue based on the average queue length, the compression target value and the second preset value.

[0084] Optionally, the compression target value refers to an ideal queue length threshold preset by the intelligent computing center switch, representing the target value for the queue to escape congestion. Based on the average queue length, the compression target value, and a second preset value, a speed reduction factor corresponding to the current queue is obtained. This factor is used to control the queue length to below the congestion threshold, thereby resolving potential network congestion.

[0085] For example, the pressure drop coefficient The calculation formula is:

[0086]

[0087] in, is a constant less than 1, K min is the congestion threshold. Used to reduce the queue length to below K min , to quickly resolve congestion, the value can usually be set to 0.95 or 0.9, which not only avoids the risk of congestion, but also makes full use of bandwidth and avoids wasting network resources.

[0088] In this embodiment, by obtaining the compression target value, the speed reduction coefficient corresponding to the current queue is obtained according to the average queue length, the compression target value and the second preset value, and the compression reduction coefficient can be accurately obtained, thereby accurately achieving network congestion control.

[0089] In an exemplary embodiment, the steps of constructing a direct return congestion notification message based on the speed reduction coefficient and the queue messages cached in the current queue include: obtaining a copy message of the queue message; sending the queue message to the receiving end according to the IP address information in the queue message; and updating the copy message based on the speed reduction coefficient to obtain a direct return congestion notification message.

[0090] Optionally, when the intelligent computing center switch detects the risk of congestion, it sends the message normally on the one hand, and on the other hand constructs a direct return congestion notification message in the name of the target receiver based on the message.

[0091] For example, under the RoCEv2 protocol, when the intelligent computing center switch detects congestion risk, it activates the DCQCN CP-side algorithm to construct a direct congestion notification message (CNP message) for intelligent computing center network congestion control. By bypassing the ECN marking process for QP packets and directly constructing CNP message packets, congestion information can be fed back more efficiently, reducing the latency that may be associated with traditional methods.

[0092] For example, a RoCEv2 message consists of multiple parts:

[0093] Ethernet header: Used for link layer addressing during transmission in Ethernet.

[0094] IP header: Contains information such as source IP address and destination IP address, and is used for routing selection at the network layer.

[0095] UDP header: includes source port and destination port, etc., and is at the transport layer.

[0096] InfiniBand Base Transport Header (BTH): This is the RDMA transport header. It contains important fields such as the operation code (Op Code) and the destination queue pair (Destination QP). It is used to control and identify RDMA operations.

[0097] Infini Band Payload: carries the actual data to be transmitted.

[0098] ICRC (Integrity Checksum): used to verify whether errors occur during message transmission.

[0099] FCS (Frame Check Sequence): checks the entire frame.

[0100] In this embodiment, by obtaining a copy of the queue message, the queue message is sent to the receiving end according to the IP address information in the queue message, and the copy message is updated based on the speed reduction coefficient to obtain a direct return congestion notification message, which can more efficiently feedback congestion information and reduce delays in the congestion control process.

[0101] In an exemplary embodiment, the copied message includes an IP packet header, a protocol data packet header, a basic transmission header and a data field; based on the speed reduction coefficient, the copied message is updated to obtain a direct return congestion notification message, including: performing address swapping on the IP packet header and port swapping on the protocol data packet header; setting the operation type field in the basic transmission header to a preset value, and storing the integer part and the decimal part of the speed reduction coefficient in two reserved fields in the basic transmission header respectively; the preset value is used to identify the message identity of the direct return congestion notification message; and deleting the data field.

[0102] For example, after copying the original message, the intelligent computing center switch swaps the destination and source IP addresses in the IP header, ensuring that the constructed CNP message is directly returned to the source sender. The source and destination ports in the UDP header are then swapped. The swapped port numbers align with the adjusted IP address, ensuring that the message is accurately returned to the sender's application. For BTH, the Op Code field is set to 0x81, and the one-bit "B" field is set to "1." The Op Code field identifies the message's operation type; a value of 0x81 indicates it's a CNP message. Setting the "B" field to "1" further clarifies its identity as a CNP message. Other fields remain unchanged, ensuring that the sender can identify basic message attributes and identifiers. Finally, the Infini Band Payload field is deleted. Since CNP messages primarily notify the sender of network congestion and do not carry actual data, removing this field reduces the size of the returned CNP message, enabling faster and more efficient network transmission and reducing network traffic burden.

[0103] For example, the deceleration coefficient needs to be passed to the source sender through the CNP message, and the deceleration coefficient is stored in the reserved fields Reserved1 and Reserved2 in the RoCEv2 message, where the Reserved1 field stores the integer part of the deceleration coefficient and the Reserved2 field stores the decimal part of the deceleration coefficient.

[0104] In this embodiment, by updating the duplicate message based on the speed reduction coefficient to obtain a direct return congestion notification message, congestion information can be fed back more efficiently, reducing delays in the congestion control process.

[0105] In an exemplary embodiment, after receiving a direct return congestion notification message, the sender parses the Op Code and B field of the BTH and determines that the message is a CNP message message. The sender parses the Destination QP and PSN fields to determine which QP queue the CNP message is notifying of impending congestion. The sender then reads the Reserved1 and Reserved2 fields from the BTH, uses the Reserved1 field as the integer part and the Reserved2 field as the decimal part, and restores the speed reduction coefficient of the corresponding QP queue's sending rate. The sender then reduces the sending rate of the QP queue in the direct return CNP message according to the speed reduction coefficient, and then discards the CNP message.

[0106] For example, the mathematical expression for reducing the transmission rate is:

[0107]

[0108] in, The current sending rate of the QP queue is used to reduce the sending rate of the QP message queue according to the above formula.

[0109] In an exemplary embodiment, the method further includes: in the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient, when the timer is in a timeout state, obtaining the average queue length corresponding to the next cycle of the current queue; when the average queue length corresponding to the next cycle meets the congestion condition, returning to the step of obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length, and continuing to execute until the average queue length of the current queue does not meet the congestion condition.

[0110] Optionally, after the sending end receives the CNP message and performs speed reduction, a timer is started. When the timer times out, it indicates that a complete feedback cycle has ended and the current queue status needs to be re-evaluated. When the timer times out, it means that the previous round of speed reduction measures has been executed for a sufficient time and it is necessary to check whether the congestion has been relieved. At this time, the intelligent computing center switch obtains the average queue length of the next cycle and performs the above-mentioned congestion control process.

[0111] In this embodiment, by obtaining the average queue length of the current queue corresponding to the next cycle of the current cycle when the timer is in the timeout state, and continuing the congestion control process, it is possible to achieve refined control of network congestion, thereby improving network stability and resource utilization.

[0112] In an exemplary embodiment, Figure 3As shown, an intelligent computing center congestion control system is provided, including: a core processing unit 302 of the CP algorithm of the intelligent computing center switch, a QP queue buffer area 304, a timer 306, a parameter register 308, a congestion control speed reduction coefficient calculation engine 310 and a CNP congestion message packet generator 312.

[0113] The core processing unit 302 of the intelligent computing center switch's CP algorithm handles congestion sensing and the CP-side functions of the DCQCN algorithm, while also coordinating and controlling the logical business processes of other units. The QP queue buffer 304 is the cache area for the intelligent computing center switch to store QP messages. The timer 306 is a timing device that controls operations related to QP queue transmission, congestion sensing, and CNP message generation. The parameter register 308 contains a series of key parameters for the CP-side DCQCN algorithm, such as the low-pass filter window parameter. These parameters can be modified by the core processing unit 302. Under the control of the core processing unit 302, the congestion control rate reduction coefficient calculation engine 310 calculates the pressure reduction rate coefficient for the QP source sender (RP side) based on the current congestion status of the QP queue and outputs the result to the CNP congestion message generator 312. Under the control of the core processing unit 302, the CNP congestion message generator 312 is used to generate a direct return CNP message corresponding to the congestion message in the QP queue buffer area 304. The generator obtains the speed reduction coefficient from the congestion control speed reduction coefficient calculation engine 310, packages it into the Reserved1 and Reserved2 fields of the CNP message, and sends it out from the corresponding sending port.

[0114] In this embodiment, by constructing a direct return CNP message packet and coordinating the source sending end to reduce the sending rate of the QP queue, the end-to-end network collaborative traffic congestion control within the intelligent computing center is implemented based on the existing RoCEv2 protocol, effectively handling the network congestion caused by Incast traffic, with a wide range of applications and easy implementation.

[0115] In an exemplary embodiment, Figure 4 As shown, a congestion control method for an intelligent computing center is provided, which includes the following steps:

[0116] S402: Obtain a queue length of a current queue corresponding to a current port in a current cycle and a previous average queue length corresponding to a cycle before the current cycle; and obtain an average queue length corresponding to the current queue in the current cycle based on the queue length, the previous average queue length, and a first preset value.

[0117] S404: When the average queue length meets the congestion condition, obtain a compression target value; the compression target value is used to represent the target value corresponding to the queue length when the congestion condition is not met; and obtain a speed reduction coefficient corresponding to the current queue based on the average queue length, the compression target value, and the second preset value.

[0118] S406: Obtain a copy of the queue message; and send the queue message to the receiving end according to the IP address information in the queue message.

[0119] The copied message includes an IP packet header, a protocol data packet header, a basic transmission header and a data field.

[0120] S408: Perform address swapping on the IP packet header and port swapping on the protocol data packet header; set the operation type field in the basic transmission header to a preset value, and store the integer part and the decimal part of the speed reduction coefficient in two reserved fields in the basic transmission header respectively; the preset value is used to identify the message identity of the direct return congestion notification message; delete the data field.

[0121] S410: Sending the direct return congestion notification message to the sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient, and also instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

[0122] S412: In the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient, when the timer is in the timeout state, obtain the average queue length of the current queue corresponding to the next cycle of the current cycle; when the average queue length corresponding to the next cycle meets the congestion condition, return to the step of obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length, and continue to execute until the average queue length of the current queue does not meet the congestion condition.

[0123] In this embodiment, by obtaining the average queue length of the current queue corresponding to the current port, when the average queue length meets the congestion condition, the speed reduction coefficient corresponding to the current queue is obtained according to the average queue length, and based on the speed reduction coefficient and the queue messages cached in the current queue, a direct return congestion notification message is constructed, and the direct return congestion notification message is sent to the sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient. This can efficiently and accurately transmit the speed reduction instruction and effectively solve network congestion.

[0124] It should be understood that, although the various steps in the flowcharts involved in the above embodiments are displayed in sequence according to the instructions of the arrows, these steps are not necessarily performed in sequence in the order indicated by the arrows. Unless clearly stated herein, the execution of these steps is not strictly limited in order, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of the steps or stages in other steps or other steps. It is understandable that the various steps in different embodiments can be freely combined as needed, and the various non-contradictory schemes formed by the combination all fall within the scope of protection of this application.

[0125] Based on the same inventive concept, embodiments of the present application also provide an intelligent computing center congestion control device for implementing the intelligent computing center congestion control method described above. The solution provided by this device is similar to the solution described in the method described above. Therefore, the specific limitations in one or more embodiments of the intelligent computing center congestion control device provided below can be found in the limitations of the intelligent computing center congestion control method described above and will not be repeated here.

[0126] In an exemplary embodiment, Figure 5 As shown, a congestion control device for an intelligent computing center is provided, comprising: a queue acquisition module 502, a speed reduction acquisition module 504, a message construction module 506, and a congestion notification module 508, wherein:

[0127] The queue acquisition module 502 is configured to acquire the average queue length of the current queue corresponding to the current port.

[0128] The speed reduction acquisition module 504 is configured to acquire a speed reduction coefficient corresponding to the current queue according to the average queue length when the average queue length meets the congestion condition.

[0129] The message construction module 506 is used to construct a direct return congestion notification message based on the speed reduction coefficient and the queue messages buffered in the current queue; the integer part and the fractional part of the speed reduction coefficient are respectively stored in two reserved fields of the direct return congestion notification message.

[0130] The congestion notification module 508 is used to send the direct return congestion notification message to the sending end to instruct the sending end to parse the direct return congestion notification message, obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

[0131] In an exemplary embodiment, the queue acquisition module 502 is also used to obtain the queue length of the current queue corresponding to the current port in the current cycle, and the previous average queue length corresponding to the current queue in the cycle before the current cycle; based on the queue length, the previous average queue length and the first preset value, the average queue length corresponding to the current queue in the current cycle is obtained.

[0132] In an exemplary embodiment, the speed reduction acquisition module 504 is also used to obtain a compression target value; the compression target value is used to represent the target value corresponding to the queue length when the congestion condition is not met; based on the average queue length, the compression target value and the second preset value, the speed reduction coefficient corresponding to the current queue is obtained.

[0133] In an exemplary embodiment, the message construction module 506 is also used to obtain a copy message of the queue message; send the queue message to the receiving end according to the IP address information in the queue message; update the copy message based on the speed reduction coefficient to obtain a direct return congestion notification message.

[0134] In an exemplary embodiment, the copied message includes an IP packet header, a protocol data packet header, a basic transmission header and a data field; the message construction module 506 is also used to perform address swap processing on the IP packet header and port swap processing on the protocol data packet header; the operation type field in the basic transmission header is set to a preset value, and the integer part and the decimal part of the speed reduction coefficient are respectively stored in two reserved fields in the basic transmission header; the preset value is used to identify the message identity of the direct return congestion notification message; and the data field is deleted.

[0135] In an exemplary embodiment, the message construction module 506 is also used to obtain the average queue length corresponding to the next cycle of the current cycle of the current queue when the timer is in a timeout state during the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient; when the average queue length corresponding to the next cycle meets the congestion condition, return to the step of obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length, and continue to execute until the average queue length of the current queue does not meet the congestion condition.

[0136] Each module in the aforementioned intelligent computing center congestion control device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a computer device's memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0137] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 6As shown. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and 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 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 external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means can be implemented via Wi-Fi, a mobile cellular network, near field communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a congestion control method for an intelligent computing center. The display unit of the computer device is used to form a visually visible image, and can 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 casing, or an external keyboard, touchpad or mouse.

[0138] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0139] 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 following steps when executing the computer program: obtaining an average queue length of a current queue corresponding to a current port; obtaining a speed reduction coefficient corresponding to the current queue according to the average queue length when the average queue length meets a congestion condition; constructing a direct return congestion notification message based on the speed reduction coefficient and queue messages cached in the current queue; storing the integer part and the decimal part of the speed reduction coefficient in two reserved fields of the direct return congestion notification message respectively; sending the direct return congestion notification message to a sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

[0140] In one embodiment, obtaining the average queue length of the current queue corresponding to the current port involved in executing a computer program by the processor includes: obtaining the queue length of the current queue corresponding to the current port in a current cycle, and the previous average queue length corresponding to the current queue in a cycle before the current cycle; and obtaining the average queue length corresponding to the current queue in the current cycle based on the queue length, the previous average queue length, and a first preset value.

[0141] In one embodiment, the processor executes a computer program involving obtaining a speed reduction coefficient corresponding to a current queue based on an average queue length, including: obtaining a compression target value; the compression target value is used to represent a target value corresponding to the queue length when congestion conditions are not met; and obtaining a speed reduction coefficient corresponding to the current queue based on the average queue length, the compression target value, and a second preset value.

[0142] In one embodiment, when a processor executes a computer program, it involves constructing a direct return congestion notification message based on a speed reduction coefficient and queue messages cached in a current queue, including: obtaining a copy message of the queue message; sending the queue message to a receiving end according to the IP address information in the queue message; and updating the copy message based on the speed reduction coefficient to obtain a direct return congestion notification message.

[0143] In one embodiment, the copied message includes an IP packet header, a protocol data packet header, a basic transmission header and a data field; when the processor executes the computer program, the copied message is updated based on the speed reduction coefficient to obtain a direct return congestion notification message, including: address swapping processing on the IP packet header and port swapping processing on the protocol data packet header; setting the operation type field in the basic transmission header to a preset value, and storing the integer part and the decimal part of the speed reduction coefficient in two reserved fields in the basic transmission header respectively; the preset value is used to identify the message identity of the direct return congestion notification message; and deleting the data field.

[0144] In one embodiment, when executing the computer program, the processor further implements the following steps: in the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient, when the timer is in a timeout state, obtaining the average queue length corresponding to the next cycle of the current queue; when the average queue length corresponding to the next cycle meets the congestion condition, returning to the step of obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length, and continuing to execute until the average queue length of the current queue does not meet the congestion condition.

[0145] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: obtaining the average queue length of the current queue corresponding to the current port; obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length when the average queue length meets the congestion condition; constructing a direct return congestion notification message based on the speed reduction coefficient and the queue messages cached in the current queue; storing the integer part and the decimal part of the speed reduction coefficient in two reserved fields of the direct return congestion notification message respectively; sending the direct return congestion notification message to the sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient, and also instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

[0146] In one embodiment, the computer program, when executed by a processor, involves obtaining an average queue length of a current queue corresponding to a current port, including: obtaining a queue length of the current queue corresponding to the current port in a current cycle, and a previous average queue length corresponding to the current queue in a cycle previous to the current cycle; and obtaining an average queue length corresponding to the current queue in the current cycle based on the queue length, the previous average queue length, and a first preset value.

[0147] In one embodiment, when a computer program is executed by a processor, the method involves obtaining a speed reduction coefficient corresponding to a current queue based on an average queue length, including: obtaining a compression target value; the compression target value is used to represent the target value corresponding to the queue length when the congestion condition is not met; and obtaining a speed reduction coefficient corresponding to the current queue based on the average queue length, the compression target value, and a second preset value.

[0148] In one embodiment, when a computer program is executed by a processor, it involves constructing a direct return congestion notification message based on a speed reduction coefficient and queue messages cached in a current queue, including: obtaining a copy message of the queue message; sending the queue message to a receiving end according to the IP address information in the queue message; and updating the copy message based on the speed reduction coefficient to obtain a direct return congestion notification message.

[0149] In one embodiment, the copied message includes an IP packet header, a protocol data packet header, a basic transmission header and a data field; when the computer program is executed by the processor, the copied message is updated based on the speed reduction coefficient to obtain a direct return congestion notification message, including: address swapping processing on the IP packet header and port swapping processing on the protocol data packet header; setting the operation type field in the basic transmission header to a preset value, and storing the integer part and the decimal part of the speed reduction coefficient in two reserved fields in the basic transmission header respectively; the preset value is used to identify the message identity of the direct return congestion notification message; and deleting the data field.

[0150] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient, when the timer is in a timeout state, the average queue length corresponding to the current queue in the next cycle of the current cycle is obtained; when the average queue length corresponding to the next cycle meets the congestion condition, the step of obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length is returned, and the execution is continued until the average queue length of the current queue does not meet the congestion condition.

[0151] In one embodiment, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the following steps: obtaining an average queue length of a current queue corresponding to a current port; obtaining a speed reduction coefficient corresponding to the current queue based on the average queue length when the average queue length satisfies a congestion condition; constructing a direct return congestion notification message based on the speed reduction coefficient and queue messages cached in the current queue; storing the integer part and the decimal part of the speed reduction coefficient in two reserved fields of the direct return congestion notification message, respectively; and sending the direct return congestion notification message to a sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue based on the speed reduction coefficient.

[0152] In one embodiment, the computer program, when executed by a processor, involves obtaining an average queue length of a current queue corresponding to a current port, including: obtaining a queue length of the current queue corresponding to the current port in a current cycle, and a previous average queue length corresponding to the current queue in a cycle previous to the current cycle; and obtaining an average queue length corresponding to the current queue in the current cycle based on the queue length, the previous average queue length, and a first preset value.

[0153] In one embodiment, when a computer program is executed by a processor, the method involves obtaining a speed reduction coefficient corresponding to a current queue based on an average queue length, including: obtaining a compression target value; the compression target value is used to represent the target value corresponding to the queue length when the congestion condition is not met; and obtaining a speed reduction coefficient corresponding to the current queue based on the average queue length, the compression target value, and a second preset value.

[0154] In one embodiment, when a computer program is executed by a processor, it involves constructing a direct return congestion notification message based on a speed reduction coefficient and queue messages cached in a current queue, including: obtaining a copy message of the queue message; sending the queue message to a receiving end according to the IP address information in the queue message; and updating the copy message based on the speed reduction coefficient to obtain a direct return congestion notification message.

[0155] In one embodiment, the copied message includes an IP packet header, a protocol data packet header, a basic transmission header and a data field; when the computer program is executed by the processor, the copied message is updated based on the speed reduction coefficient to obtain a direct return congestion notification message, including: address swapping processing on the IP packet header and port swapping processing on the protocol data packet header; setting the operation type field in the basic transmission header to a preset value, and storing the integer part and the decimal part of the speed reduction coefficient in two reserved fields in the basic transmission header respectively; the preset value is used to identify the message identity of the direct return congestion notification message; and deleting the data field.

[0156] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: in the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient, when the timer is in a timeout state, the average queue length corresponding to the current queue in the next cycle of the current cycle is obtained; when the average queue length corresponding to the next cycle meets the congestion condition, the step of obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length is returned, and the execution is continued until the average queue length of the current queue does not meet the congestion condition.

[0157] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. 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. In particular, any reference to memory, database, or other media used in the embodiments provided in this 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. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0158] The technical features of the above embodiments can be combined arbitrarily. In order 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.

[0159] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A congestion control method for an intelligent computing center, characterized in that: Applied to an intelligent computing center switch; the method includes: Get the average queue length of the current queue corresponding to the current port; When the average queue length meets the congestion condition, obtaining a speed reduction coefficient corresponding to the current queue according to the average queue length; Constructing a direct return congestion notification message based on the speed reduction coefficient and the queue messages cached in the current queue; storing the integer part and the fractional part of the speed reduction coefficient in two reserved fields of the direct return congestion notification message respectively; The direct return congestion notification message is sent to the sending end to instruct the sending end to parse the direct return congestion notification message to obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

2. The method according to claim 1, characterized in that Obtaining the average queue length of the current queue corresponding to the current port includes: Obtaining a queue length of a current queue corresponding to a current port in a current cycle and a previous average queue length of the current queue in a cycle previous to the current cycle; The average queue length corresponding to the current queue in the current period is obtained according to the queue length, the previous average queue length and a first preset value.

3. The method according to claim 1, characterized in that The obtaining, according to the average queue length, a speed reduction coefficient corresponding to the current queue includes: Obtaining a compression target value; the compression target value is used to represent a target value corresponding to the queue length when the congestion condition is not met; A speed reduction coefficient corresponding to the current queue is obtained according to the average queue length, the compression target value, and a second preset value.

4. The method according to claim 1, wherein The constructing of a direct return congestion notification message based on the speed reduction coefficient and the queue messages cached in the current queue includes: Obtain a copy of the queue message; Sending the queue message to the receiving end according to the IP address information in the queue message; Based on the speed reduction coefficient, the copied message is updated to obtain a direct return congestion notification message.

5. The method according to claim 4, characterized in that The duplicate message includes an IP header, a protocol data header, a basic transmission header, and a data field; and the updating process of the duplicate message based on the speed reduction coefficient to obtain a direct return congestion notification message includes: Performing address swapping processing on the IP packet header and port swapping processing on the protocol data packet header; Setting the operation type field in the basic transmission header to a preset value, and storing the integer part and the fractional part of the speed reduction coefficient in two reserved fields in the basic transmission header respectively; the preset value is used to identify the message identity of the direct return congestion notification message; Delete the data field.

6. The method according to claim 2, characterized in that The method further comprises: In the process of instructing the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient, when the timer is in a timeout state, obtaining an average queue length of the current queue corresponding to a next cycle of the current cycle; If the average queue length corresponding to the next cycle meets the congestion condition, return to the step of obtaining the speed reduction coefficient corresponding to the current queue according to the average queue length, and continue to execute until the average queue length of the current queue does not meet the congestion condition.

7. A congestion control device for an intelligent computing center, characterized in that: The device comprises: The queue acquisition module is used to obtain the average queue length of the current queue corresponding to the current port; a speed reduction acquisition module, configured to obtain a speed reduction coefficient corresponding to the current queue according to the average queue length when the average queue length meets the congestion condition; a message construction module, configured to construct a direct return congestion notification message based on the speed reduction coefficient and queue messages cached in the current queue; the integer part and the fractional part of the speed reduction coefficient are respectively stored in two reserved fields of the direct return congestion notification message; The congestion notification module is used to send the direct return congestion notification message to the sending end to instruct the sending end to parse the direct return congestion notification message, obtain the current queue and the speed reduction coefficient, and also instruct the sending end to adjust the sending rate of the current queue according to the speed reduction coefficient.

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