WO2020207479A1 - Method and device for controlling network congestion - Google Patents

Abstract

The embodiments of the present application disclose a method and device for controlling network congestion, relate to the field of communication technology, and solve the problem of inaccurate RTT measured in the prior art, which causes the congestion depth of the network queue to not be effectively controlled. The specific solution is: applied to a first apparatus, the first apparatus is an apparatus that sends the data message, the first apparatus sends a first message to a second apparatus, the first message carries a first timestamp; the first timestamp is the local timestamp when the first message is sent; the first apparatus receives a second message sent by the second apparatus, the second message carries the first timestamp; subtracting the first timestamp from the second timestamp to obtain a first RTT; adjusting the sending rate of the data message according to the first RTT; wherein, the priority of the first message is the same as the priority of the data message, the priority of the second message is higher than the priority of the data message.

Description

Method and device for controlling network congestion

This application claims the priority of a Chinese patent application filed with the State Intellectual Property Office on April 12, 2019, the application number is 201910295531.0, and the application name is "a network congestion control method and device", the entire content of which is incorporated herein by reference Applying.

Technical field

The embodiments of the present application relate to the field of communication technologies, and in particular, to a method and device for network congestion control.

Background technique

At present, computers in the data center network can exchange data through remote direct memory access (RDMA), so that the computer's network interface card (NIC) can read or read from the memory of another computer through the network. Write data to the memory of another computer without the intervention of the computer's operating system. RDMA running on Ethernet is called RDMA over Converged Ethernet (RoCE) technology.

In order to avoid network congestion and packet loss from causing performance loss to RoCE, an existing flow control method measures the round trip delay of a packet (64KB), calculates the round trip time (Round Trip Time, RTT), and calculates the round trip time according to the round trip time. RTT adjusts the sending rate. As shown in Figure 1, the calculation of RTT in the prior art is performed on a data segment. The size of the data segment is 64KB and can contain multiple messages. Host A is sending the first message of this data segment. When the message is sent, the time stamp t _s at the time of sending is recorded, and when the host A receives the Acknowledgement (ACK) sent by the host B, the completion time t _{comp is} recorded. As shown in Figure 1, the round-trip time RTT = t _comp- t _s- t _serial , where t _serial is the serialized transmission time of the data segment, and the serialized transmission time of the data segment is divided by the data segment size (64KB) At the line rate.

However, the size of the data segment in this solution is 64KB, and the size of the data block required by each request in actual applications is not fixed, so there is no guarantee of an ACK every 64KB. If the data block is small, there will be multiple data segments per 64KB ACK, using different ACK completion times to calculate RTT will affect the accuracy of RTT, so the RTT calculated by this scheme is not accurate, resulting in the congestion depth of the network queue cannot be effectively controlled; and the RTT measured by this scheme is reversed The influence of path congestion cannot accurately reflect whether congestion occurs in the request direction or the response direction, which may cause misjudgment by the control system.

Summary of the invention

The embodiments of the present application provide a network congestion control method and device, which can avoid the influence of reverse path congestion, accurately control the congestion depth of network queues, and improve system performance.

In order to achieve the foregoing objectives, the following technical solutions are adopted in the embodiments of this application:

In a first aspect of the embodiments of the present application, a network congestion control method is provided. The method is applied to a first device. The first device is a device that sends a data packet. The method includes: the first device sends a first device to a second device. A message, the first message carries a first time stamp; the first time stamp is a local time stamp when the first message is sent; the first device receives the second message sent by the second device, The second message carries the first time stamp; the first time stamp is subtracted from the second time stamp to obtain the first round-trip time RTT; the second time stamp is when the first device receives the second message According to the first RTT, adjust the sending rate of the data message; wherein the priority of the first message is the same as the priority of the data message, and the priority of the second message is higher than the priority The priority of the data message. Based on this solution, the measurement of the first RTT is not affected by whether the reverse path (the transmission direction of the untransmitted service message) is congested, and the determined first RTT is more accurate. Therefore, the first RTT is used to adjust the data message When sending rate, it can reduce network queue congestion and improve system performance. It is understandable that the first RTT not only considers the queuing and processing time in the buffer of the switch (or router), but also avoids the influence of congestion in the reverse path (the transmission direction of untransmitted service packets). Therefore, the first RTT Related to the degree of network congestion, it dynamically changes with changes in the degree of network congestion, and can more accurately reflect the degree of congestion of the current network. Therefore, the first RTT can be called a dynamic RTT.

With reference to the first aspect, in a possible implementation manner, the above method further includes: the first device sends a third message to the second device, the third message carries a third time stamp, and the third time The stamp is the local timestamp when the third message is sent; the priority of the third message is higher than the priority of the data message; the first device receives the fourth message sent by the second device; the fourth The message carries the third time stamp; the priority of the fourth message is higher than the priority of the data message; the fourth time stamp is subtracted from the third time stamp to obtain the second RTT; the fourth time The stamp is the local timestamp when the first device receives the fourth message. Based on this solution, the second RTT can be measured more accurately by using the third message and the fourth message with a priority higher than the priority of the data message. It is understandable that under the condition that the data transmission path between the network card of the first device and the network card of the second device is unchanged, the value of the second RTT is basically fixed, and may slightly change with network performance. Therefore, the second RTT can be called a fixed RTT.

Combining the first aspect and the foregoing possible implementation manners, in another possible implementation manner, the foregoing adjusting the sending rate of the data message according to the foregoing first RTT includes: subtracting the foregoing second RTT from the foregoing first RTT, Obtain the time difference; the time difference is used to indicate the congestion depth of the network queue; according to the time difference, adjust the sending rate of the data message. Based on this solution, the time difference obtained by the difference between the first RTT and the second RTT can accurately reflect the depth of network queue congestion between the first device and the second device. Therefore, when adjusting the sending rate of data packets according to the time difference, it can be Effectively reduce the congestion depth of network queues and improve system performance.

Combining the first aspect and the foregoing possible implementation manners, in another possible implementation manner, the foregoing adjusting the sending rate of data packets according to the foregoing time difference includes: if the time difference is less than a first preset threshold, increasing the foregoing data The message sending rate; if the time difference is greater than the second preset threshold, the sending rate of the data message is reduced; the first preset threshold is less than the second preset threshold. Based on this solution, the sending rate of data packets can be reduced when the network queue is relatively congested, so as to reduce the congestion depth of the network queue.

Combining the first aspect and the foregoing possible implementation manners, in another possible implementation manner, the foregoing method further includes: if the first device determines that the first preamble has been sent cumulatively since the last time the first message was sent. Set the number of data packets to obtain the third RTT; or, if the first device determines that the time interval between the current time and the last sending of the first packet reaches the first preset duration, obtain the third RTT, and Record the current timestamp. It should be noted that the third RTT and the foregoing first RTT are dynamic RTTs at different moments. Based on this solution, since the depth of network congestion changes dynamically, the dynamic RTT can be detected periodically to obtain the current network congestion degree.

Combining the first aspect and the foregoing possible implementation manners, in another possible implementation manner, the foregoing method further includes: if the first device determines that the second preset has been sent cumulatively since the last time the third message was sent. Set the number of data packets to obtain the fourth RTT; or, if the first device determines that the time interval between the current time and the last sending of the third packet reaches the second preset duration, obtain the fourth RTT, and Record the current timestamp. It should be noted that the fourth RTT and the foregoing second RTT are fixed RTTs at different times. Based on this solution, the fixed RTT can be periodically and cyclically detected, so that when the data transmission path between the first device and the second device changes, the fixed RTT corresponding to the new transmission path can be detected more accurately.

Combining the first aspect and the foregoing possible implementation manners, in another possible implementation manner, the foregoing third time stamp is carried in a reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried in The payload of the message RDMA. Based on this solution, by using the reserved field of the existing protocol to carry the timestamp, compared with the prior art, there is no need to record the relationship between the timestamp and the message sequence number and occupy less resources.

Combining the first aspect and the foregoing possible implementation manners, in another possible implementation manner, if the foregoing first message is the foregoing data message, the foregoing first time stamp is carried in a reserved field in the BTH of the RDMA message ; If the first message is different from the data message, the first time stamp is carried in the reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried in the payload of the message RDMA. . Based on this solution, by using the reserved field of the existing protocol to carry the timestamp, compared with the prior art, there is no need to record the relationship between the timestamp and the message sequence number and occupy less resources.

A second aspect of the embodiments of the present application provides a network congestion control method, the method includes: a second device receives a first message sent by a first device, the first message carries a first time stamp; The timestamp is the local timestamp when the first message is sent; the first device is the device that sends the data message; the second device sends a second message to the first device, and the second message carries the first message. A timestamp; where the priority of the first message is the same as the priority of the data message, and the priority of the second message is higher than the priority of the data message. Based on this solution, by using a message with a higher priority than that of a data message in the transmission direction of the unsent data, the first RTT measured by the first device is not affected by the reverse path (the transmission of the untransmitted service message). Transmission direction) is the impact of congestion.

With reference to the second aspect, in a possible implementation manner, the above method further includes: the second device receives a third message sent by the first device, the third message carries a third timestamp, and the third The timestamp is the local timestamp when the third message is sent; the priority of the third message is higher than the priority of the data message; the second device sends the fourth message to the first device; The fourth message carries the foregoing third time stamp; the priority of the fourth message is higher than the priority of the foregoing data message. Based on this solution, the second RTT measured by the first device is more accurate by using the third message and the fourth message whose priority is higher than the priority of the data message.

In a third aspect of the embodiments of the present application, a network congestion control device is provided. The device is a device for sending data packets. The device includes: a processing unit and a transceiver unit; the processing unit is configured to: Send a first message to the second device, where the first message carries a first time stamp; the first time stamp is the local time stamp when the first message is sent; The second message sent by the second device, the second message carries the first timestamp; subtract the first timestamp from the second timestamp to obtain the first round-trip time RTT; The second time stamp is the local time stamp when the device receives the second message; the sending rate of the data message is adjusted according to the first RTT; where the priority of the first message and the data The priorities of the messages are the same, and the priority of the second message is higher than the priority of the data message.

With reference to the third aspect, in a possible implementation manner, the processing unit is further configured to: send a third message to the second device through the transceiver unit, the third message carrying a third message Time stamp, the third time stamp is the local time stamp when the third message is sent; the priority of the third message is higher than the priority of the data message; The fourth message sent by the second device; the fourth message carries the third timestamp; the priority of the fourth message is higher than the priority of the data message; the fourth time is used Subtract the third time stamp from the stamp to obtain the second RTT; the fourth time stamp is the local time stamp when the device receives the fourth message.

In combination with the third aspect and the foregoing possible implementation manners, in another possible implementation manner, the processing unit is specifically configured to: subtract the second RTT from the first RTT to obtain the time difference; Used to indicate the depth of network queue congestion; adjust the sending rate of data packets according to the time difference.

Combining the third aspect and the foregoing possible implementation manners, in another possible implementation manner, the processing unit is specifically configured to: if the time difference is less than a first preset threshold, increase the sending of the data message Rate; if the time difference is greater than a second preset threshold, reduce the sending rate of the data message; the first preset threshold is less than the second preset threshold.

In combination with the third aspect and the foregoing possible implementation manners, in another possible implementation manner, the processing unit is further configured to: if the processing unit determines that the first message has been sent cumulatively since the last time the first message was sent The first preset number of data packets, obtain the third RTT; or, if the processing unit determines that the time interval between the current time and the last time the first packet is sent reaches the first preset duration, obtain the third RTT , And record the current timestamp.

In combination with the third aspect and the foregoing possible implementation manners, in another possible implementation manner, the processing unit is further configured to: if the processing unit determines that the third message has been sent cumulatively since the last time the third message was sent The second preset number of data packets, obtain the fourth RTT; or, if the processing unit determines that the time interval between the current time and the last sending of the third packet reaches the second preset duration, obtain the fourth RTT , And record the current timestamp.

Combining the third aspect and the foregoing possible implementation manners, in another possible implementation manner, the third timestamp is carried in a reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried In the payload of the message RDMA.

Combining the third aspect and the foregoing possible implementation manners, in another possible implementation manner, if the first message is the data message, the first time stamp is carried in the BTH of the RDMA message In the reserved field; if the first message and the data message are different, the first time stamp is carried in the reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried in the message In the payload of the text RDMA.

In a fourth aspect of the embodiments of the present application, there is provided a network congestion control device, which includes: a processing unit and a transceiving unit; the processing unit is configured to receive a first message sent by a first device through the transceiving unit , The first message carries a first timestamp; the first timestamp is a local timestamp when the first message is sent; the first device is a device that sends a data message; through the transceiver unit Send a second message to the first device, and the second message carries the first time stamp; wherein the priority of the first message is the same as the priority of the data message, so The priority of the second message is higher than the priority of the data message; or, the priority of the first message is higher than the priority of the data message, and the priority of the second message It is the same as the priority of the data message.

With reference to the fourth aspect, in a possible implementation manner, the processing unit is further configured to: receive, through the transceiver unit, a third message sent by the first device, where the third message carries a Three timestamps, where the third timestamp is the local timestamp when the third message is sent; the priority of the third message is higher than the priority of the data message; The first device sends a fourth message; the fourth message carries the third timestamp; the priority of the fourth message is higher than the priority of the data message.

For the description of the effects of the foregoing third aspect and various implementation manners of the third aspect, reference may be made to the description of the corresponding effects of the first aspect and the various implementation manners of the first aspect. The foregoing fourth aspect and the various implementation manners of the fourth aspect For the effect description, reference may be made to the second aspect and the description of the corresponding effects of the various implementation manners of the second aspect, which are not repeated here.

In a fifth aspect of the embodiments of the present application, a computer storage medium is provided, and computer program code is stored in the computer storage medium. When the computer program code runs on a processor, the processor executes any of the above The network congestion control method described in the aspect.

The sixth aspect of the embodiments of the present application provides a computer program product that stores computer software instructions executed by the above-mentioned processor, and the computer software instructions include a program for executing the solution described in the above-mentioned aspect.

A seventh aspect of the embodiments of the present application provides a network congestion control device, which includes a transceiver, a processor, and a memory. The transceiver is used for sending and receiving information or communicating with other network elements; and the memory is used for Computer-executable instructions are stored; the processor is used to execute the computer-executed instructions to implement the network congestion control method described in any of the above aspects.

The eighth aspect of the embodiments of the present application provides a network congestion control device. The device exists in the form of a chip. The structure of the device includes a processor and a memory. The memory is used for coupling with the processor and storing the device. Necessary program instructions and data, the processor is used to execute the program instructions stored in the memory, so that the device executes the functions of the device in the above method.

Description of the drawings

Figure 1 is a schematic diagram of a network congestion control solution provided by the prior art of this application;

Figure 2 is a schematic diagram of a network architecture provided by an embodiment of the application;

FIG. 3 is a schematic flowchart of a network congestion control method provided by an embodiment of this application;

FIG. 4 is a schematic diagram of a way to carry a timestamp according to an embodiment of the application;

FIG. 5 is a schematic diagram of another way of carrying time stamps according to an embodiment of the application;

6 is a schematic flowchart of another network congestion control method provided by an embodiment of this application;

FIG. 7 is a schematic flowchart of another network congestion control method provided by an embodiment of this application;

FIG. 8 is a schematic diagram of a data transmission structure provided by an embodiment of this application;

FIG. 9 is a schematic diagram of the composition of a network congestion control device provided by an embodiment of this application;

10 is a schematic diagram of the composition of another network congestion control device provided by an embodiment of the application;

11 is a schematic diagram of the composition of another network congestion control device provided by an embodiment of the application;

FIG. 12 is a schematic diagram of the composition of another network congestion control apparatus provided by an embodiment of the application.

detailed description

The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. In this application, "at least one" refers to one or more, and "multiple" refers to two or more. "And/or" describes the association relationship of the associated objects, indicating that there can be three relationships, for example, A and/or B, which can mean: A alone exists, both A and B exist, and B exists alone, where A, B can be singular or plural. The character "/" generally indicates that the associated objects are in an "or" relationship. "The following at least one item (a)" or similar expressions refers to any combination of these items, including any combination of single item (a) or plural items (a). For example, at least one of a, b, or c can represent: a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

It should be noted that in this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or illustrations. Any embodiment or design solution described as "exemplary" or "for example" in this application should not be construed as being more preferable or advantageous than other embodiments or design solutions. To be precise, words such as "exemplary" or "for example" are used to present related concepts in a specific manner.

First, explain some terms involved in the embodiments of this application:

RDMA remote direct memory access, transfer data directly to the storage area of the computer through the network, and quickly move the data from a system to the remote system memory without any impact on the operating system, so that it does not require much computer processing functions. It can eliminate the overhead of external memory copy and context switching. The RDMA protocol enables the computer's network interface card (NIC) to read from or write data to the memory of another computer through the network without the intervention of the computer's operating system. RDMA Traversal Converged Ethernet (RoCE) is proposed by InfiniBand (IB) for RDMA to run on Ethernet. In the RoCE technology, directly bearing and running on the Ethernet link layer is called RoCE version 1 (RoCEv1), and bearing and running on the User Datagram Protocol (UDP) is called RoCE version 2 (RoCEv2).

The RTT round-trip time represents the total delay from the start of the sender sending data to the sender receiving the confirmation from the receiving end (the receiving end immediately sends the confirmation after receiving the data). RTT is determined by three parts: the propagation time of the link, the processing time of the end system, the queuing and processing time in the buffer of the switch (or router). Among them, the propagation time of the link and the processing time of the end system are relatively fixed, and the queuing and processing time in the buffer of the switch (or router) will change with the change of the congestion degree of the entire network. Therefore, the change of RTT is to a certain extent Reflects changes in the degree of network congestion.

In order to solve the problems that the round-trip time RTT measured in the prior art is inaccurate and cannot accurately reflect the specific congestion transmission direction, causing misjudgment by the control system, etc., an embodiment of the present application provides a network congestion control method that can avoid reverse paths. The impact of congestion can accurately control the depth of network queue congestion and improve system performance.

The embodiment of the application provides a network congestion control method, which is applied to a computer node in a data center that uses the RoCE protocol for data exchange. The computer node is interconnected through one or more switches; a certain topological relationship between multiple switches (For example, CLOS topology) is connected to form a data center network with one or more paths. The embodiment of the present application does not limit the topological relationship between the switches, which is only an exemplary description here.

Figure 2 is a network architecture provided by an embodiment of the application, including a computer node A and a computer node B. The computer node A can be connected to the computer node B through one or more switches. The computer node A includes a host A and a network card A. Computer node B includes host B and network card B. The network card A of the computer node A and the network card B of the computer node B exchange data through remote direct memory access (RDMA).

As shown in Figure 2, when computer node A and computer node B exchange data, a communication queue pair (Queue Pair, QP) is created. One of the communication queue pairs is a sending queue and the other is a receiving queue. QP is full-duplex communication. The end sending a request is the requesting end, and the end receiving the request and responding is the responding end. RoCE requests are issued by the application program, and the request types used for data exchange with remote computer nodes mainly include Write, Send, and Read. Among them, Write and Send are the computer requesting end to send data, and the computer responding end will respond with an acknowledgement character (Acknowledgement, ACK) after receiving the data; Read is the computer requesting end sending a read request, and the computer responding end receiving the request and responding to the read data. That is, Write/Send is the requester to send data, and Read is the responder to send data, so the data transmission directions of Write/Send and Read are different.

Exemplarily, for each QP, the data exchange is bidirectional. When network card A sends a request, network card A is the requesting end, and network card B is the responding end. At this time, RDMA Write/Send carries data on network card A, and RDMA Read carries data on network card B; when network card B sends a request, network card B is the requester, and network card A is the responder. At this time, RDMA Write/Send network card B carries data, and RDMA Read carries data on network card A. In order to accurately reflect the specific transmission direction where congestion occurs, the network congestion control method provided in the embodiment of the present application distinguishes the data transmission direction.

With reference to FIG. 2, as shown in FIG. 3, an embodiment of the present application provides a network congestion control method, which is applied to a first device, and the first device is a device that sends a data message. The method may include steps S301-S307. When the first device sends a data message, the network congestion control method of steps S301-S307 can be executed to reduce the congestion depth of the network queue.

S301. The first device sends a first message to the second device.

The first message carries the first time stamp. The first time stamp is a local time stamp when the first device sends the first message.

Exemplarily, the priority of the first message is the same as the priority of the data message. The first message may be a data message, or may be a message specifically for measuring delay, which is not limited in the embodiment of the present application. For example, when the first message is a message dedicated to measuring delay, the first message and the data message sent by the first device are different messages, but the first message and the data message have priority The same level. It can be understood that the first device in this embodiment is a device that sends a data packet, and the transmission direction of the data packet is from the first device to the second device.

Exemplarily, take the first device as the computer node A and the second device as the computer node B as an example. When computer node A sends a Write or Send request message (for example, the first message is a Write or Send request message), the Write or Send request message carries data, and the computer node A is the device that sends the data message , So the transmission direction of the data message is from computer node A to computer node B.

Exemplarily, take the first device as the computer node B and the second device as the computer node A as an example. When computer node A sends a Read request, after receiving the Read request from computer node A, computer node B sends the above-mentioned first message to computer node A, that is, computer node B is the device that sends the data message, so the data message is The transmission direction is from computer node B to computer node A. It can be understood that, in this case, optionally, before the above step S301, it may also include the second device sending a Read request to the first device.

Exemplarily, when the first message is a message specifically for measuring the delay sent by the first device, the first message carries the foregoing first time stamp. The carrying manner of the first time stamp is shown in FIG. 4 or FIG. 5.

As shown in FIG. 4, the first time stamp in the first message may be carried in a reserved field (reserved, rsvd) in the basic transport header (Base Transport Header, BTH) of the remote direct memory access RDMA of the message. There are two reserved fields in the BTH, namely the first reserved field (rsvd) and the second reserved field (rsvd). among them. The first reserved field (rsvd) is the 5th byte in the BTH, and the second reserved field (rsvd) is the lower 7 bits of the 9th byte in the BTH. The first time stamp can be carried in one of the first reserved field (rsvd) and the second reserved field (rsvd), or can be carried in a combination of two reserved fields, which is not carried out in this embodiment of the application. limited.

The operation code (Operation Code, Opcode) in Figure 4 is used to indicate the type of the data packet or the higher-level protocol type in IB PayLoad; the request event identifier (Solicited Event, SE) indicates that the responder should generate an event; the migration status The identifier (MigReq, M) is used to identify the migration status; the number of payload padding bytes (Pad Count, Pad), which identifies how many extra bytes are filled in IB PayLoad; the transport header version number (Transport Header Version, TVer), Used to indicate the version number of the packet; Partition Key is used to characterize the logical memory partition associated with this Packet; Destination Queue Pair indicates the destination serial number; A (Acknowledge Request, A) ) Request to respond with a response; Packet Sequence Number (PSN), used to detect lost or duplicate data packets.

As shown in Figure 5, the first time stamp in the above first message can also be carried in the payload of the RDMA message, and the payload part may not carry valid data; by carrying the first time stamp in the payload, there can be more Large storage space, so it can carry a more accurate time stamp. The meanings of other fields in FIG. 5 are the same as those in FIG. 4, and will not be repeated here.

Exemplarily, when the first message is a data message, the foregoing first time stamp is carried in the data message. The manner of carrying the first time stamp in the data message is shown in FIG. 4, that is, the first time stamp in the second message can be carried in a reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message. For details, please refer to the aforementioned related description, which will not be repeated here.

It is understandable that this embodiment uses the reserved field of the existing protocol to carry the time stamp. Compared with the prior art, there is no need to record the relationship between the time stamp and the message sequence number, and it takes up less resources.

S302. The second device receives the first message.

S303. The second device constructs a second message according to the first message.

Exemplarily, the second device extracts the first time stamp from the first message, and constructs a second message, and the second message carries the first time stamp.

Exemplarily, the manner of carrying the first time stamp in the second message is shown in Figure 4 or Figure 5, that is, the first time stamp in the second message can be carried in the basic transmission of remote direct memory access RDMA of the message. In the reserved field in the header BTH, or carried in the payload of the message RDMA. For details, please refer to the aforementioned related description, which will not be repeated here.

S304. The second device sends a second message to the first device.

The second message carries the foregoing first time stamp.

Exemplarily, the priority of the second message is higher than the priority of the data message. For example, the second message may be an out-of-band message. The out-of-band message is an out-of-service message and is used to assist in measuring RTT. The second message is completely independent from the service and can be sent through the underlying control module. . It is understandable that because the priority of the second message sent by the second device to the first device is higher than the priority of the data message, when sending the second message, it can be sent before the data message, so that It is not affected by whether the network is congested in the transmission direction from the second device to the first device, etc., so that the measured RTT can more accurately reflect the degree of queue congestion in the transmission direction from the first device to the second device.

Exemplarily, in this embodiment, the first device sends the first packet with the same priority as the data packet to the second device, and the second device sends the first packet with the priority higher than the priority of the data packet to the first device. The second message can avoid the influence of congestion on the reverse path (the transmission direction from the second device to the first device), and the RTT can be measured more accurately.

For example, take the first device as the computer node A and the second device as the computer node B as an example. When computer node A sends the first message (Write or Send request message) to computer node B, the priority of the first message is the same as the priority of the data message. After computer node B receives the first message, Send the second message with priority higher than the priority of the data message to computer node A, so as to avoid the influence of network congestion in the transmission direction from computer node B to computer node A, so that the measured RTT can be more accurately reflected The degree of queue congestion in the transmission direction from computer node A to computer node B.

For example, take the first device as the computer node B and the second device as the computer node A as an example. Computer node B sends a first message to computer node A (this first message is the first message that computer node B replies with the same priority as the data message after receiving the Read request sent by computer node A, the first message A message can carry data, or it can be a message with the same priority as the data message that specifically measures the delay), the first message has the same priority as the data message, and the computer node A receives After the first message, a second message with a priority higher than the priority of the data message is sent to the computer node B, so as to avoid the influence of network congestion in the transmission direction from the computer node A to the computer node B, so that the measurement The RTT can more accurately reflect the degree of queue congestion in the transmission direction from computer node B to computer node A.

It is understandable that this embodiment can avoid congestion in the reverse path (transmission direction of untransmitted service packets) by adopting a message with a higher priority than that of the data message in the transmission direction of the untransmitted service message. Influence, so that the measured delay can more accurately reflect the degree of queue congestion in the direction of transmitting service packets.

S305. The first device receives the second message.

S306. The first device subtracts the first time stamp from the second time stamp to obtain the first RTT.

The second time stamp is a local time stamp when the first device receives the second message.

Exemplarily, the second timestamp minus the first timestamp can be understood as the total experience from the first device sending the first message to the first device receiving the second message sent by the peer (the second device) Time is the first RTT.

It is understandable that since the priority of the first message is the same as the priority of the data message, and the priority of the second message is higher than the priority of the data message, the first RTT not only considers the switch (or router) The queuing and processing time in the cache, and avoids the impact of reverse path (transmission direction of untransmitted service packets) congestion, so the first RTT is related to the degree of network congestion, and will change dynamically with the degree of network congestion Changes can more accurately reflect the current network congestion. This first RTT may be referred to as dynamic RTT.

S307: The first device adjusts the sending rate of the data packet according to the first RTT.

Exemplarily, the first RTT can reflect the current congestion degree of the network queue. In the case that the data transmission path between the first device and the second device remains unchanged, the larger the first RTT, the more congested the network queue, so the sending rate of the data packet can be adjusted according to the first RTT.

For example, take the first device as the computer node A and the second device as the computer node B as an example. If the computer node A sends a Write or Send request message (the first message), the computer node A is the device that sends the data message. At this time, the transmission direction of the data message is from the computer node A to the computer node B. An RTT can reflect the congestion degree of the network queue in the transmission direction from the computer node A to the computer node B. Therefore, the computer node A can adjust the sending rate of the data message according to the first RTT.

For example, take the first device as the computer node B and the second device as the computer node A as an example. If computer node A sends a Read request, after receiving the Read request from computer node A, computer node B sends the above-mentioned first message to computer node A. This computer node B is the device that sends the data message. The transmission direction is from the computer node B to the computer node A, so the computer node B can adjust the sending rate of the data message according to the first RTT.

Exemplarily, the foregoing first device may adjust the sending rate of the data message according to the first RTT, which may include: if the first RTT is greater than the first preset threshold, it is determined that the network is relatively congested at the time, and the sending of data messages can be reduced. If the first RTT is less than the second preset threshold, it is determined that the current network is not congested, and the sending rate of data packets can be appropriately increased to make full use of the network capacity. The second preset threshold is less than or Equal to the first preset threshold. It should be noted that the embodiment of the present application does not limit the specific method of how the first device adjusts the sending rate of the data packet according to the first RTT, and is only an exemplary description here.

It should be noted that under the condition that the data transmission path between the first device and the second device is unchanged, the method of steps S301-S307 can be used to adjust the sending rate of the data packet to reduce the congestion degree of the network queue.

In the network congestion control method provided by the embodiments of the present application, the first device sends the first message to the second device through the first device; the second device receives the first message; the second device constructs the second message according to the first message; The second device sends the second message to the first device; the first device receives the second message; the first device subtracts the first timestamp from the second timestamp to obtain the first round-trip time RTT; the first device according to the first RTT To adjust the sending rate of data packets. In this embodiment, the measurement of the first RTT is not affected by the congestion of the reverse path (the transmission direction of the untransmitted service packet), and the determined first RTT is more accurate. Therefore, the transmission rate of the data packet is adjusted through the first RTT. At this time, it can reduce the network queue congestion and improve system performance.

The embodiment of the present application also provides a network congestion control method. As shown in FIG. 6, before the above step S307, the method further includes steps S601-S606.

S601. The first device sends a third packet to the second device.

The third message carries a third timestamp, and the third timestamp is a local timestamp when the third message is sent. The priority of the third message is higher than the priority of the data message. Exemplarily, the third message may be a message that does not carry data. For example, the third message may be an out-of-band message to assist in measuring RTT.

Exemplarily, the way of carrying the third time stamp in the third message is shown in Figure 4 or Figure 5, that is, the third time stamp in the third message can be carried in the basic transmission of remote direct memory access RDMA of the message. In the reserved field in the header BTH, or carried in the payload of the message RDMA. For details, please refer to the aforementioned related description, which will not be repeated here.

S602. The second device receives the third message.

S603. The second device constructs a fourth message according to the third message.

Exemplarily, the second device extracts the third timestamp from the third message, and constructs a fourth message, and the fourth message carries the third timestamp.

Exemplarily, the manner of carrying the third time stamp in the fourth message is shown in FIG. 4 or FIG. 5. For details, please refer to the aforementioned related description, which will not be repeated here.

S604: The second device sends a fourth packet to the first device.

The priority of the fourth message is higher than the priority of the data message. Exemplarily, the fourth message may be a message that does not carry data. For example, the third message may be an out-of-band message to assist in measuring RTT.

S605: The first device receives the fourth packet.

S606. The first device subtracts the third time stamp from the fourth time stamp to obtain the second RTT.

The fourth time stamp is a local time stamp when the first device receives the fourth message.

Exemplarily, the fourth timestamp minus the third timestamp can be understood as the total elapsed time from the first device sending the third message to the first device receiving the fourth message sent by the second device, which is The second RTT.

(Optional) The first device may save the second RTT in the context.

It can be understood that, in this embodiment, by sending a third message with a higher priority than a data message, and receiving a fourth message with a higher priority than the data message sent by the opposite end, it is determined that the second RTT is not affected by Whether the network is congested, the RTT between the first device and the second device can be measured more accurately.

It should be noted that under the condition that the data transmission path between the network card of the first device and the network card of the second device is unchanged, the value of the second RTT is basically fixed, and may vary slightly with network performance, etc. . This second RTT may be referred to as a fixed RTT.

It can be understood that the path for obtaining the first RTT in steps S301-S306 is the same as the path for obtaining the second RTT in steps S601-S606. The above steps S301-S306 may be executed before steps S601-S606, or may be executed after steps S601-S606, or may also be executed simultaneously with steps S601-S606, which is not limited in the embodiment of the present application.

After performing the above steps S301-S306 and S601-S606, correspondingly, in the above S307, the first device adjusts the sending rate of the data packet according to the first RTT, including: the first device adjusts according to the first RTT and the second RTT The sending rate of data packets.

Exemplarily, the first device adjusts the sending rate of the data packet according to the first RTT and the second RTT, including: the first device subtracts the second RTT from the first RTT to obtain a time difference, which is used to indicate network queue congestion Depth: The first device adjusts the sending rate of the data message according to the time difference.

(Optional) The first device may obtain the saved second RTT from the context, and subtract the second RTT saved in the context from the first RTT to obtain the time difference.

Exemplarily, the time difference between the first RTT and the second RTT may be used to indicate the depth of network queue congestion in the transmission direction from the first device to the second device. It can be understood that if the first device is a device that sends data packets, the time difference specifically represents the depth of queue congestion in the transmission direction from the first device to the second device; the smaller the time difference, the smaller the time difference, the The queue in the transmission direction to the second device is less congested; the larger the time difference is, the deeper the queue congestion in the transmission direction from the first device to the second device, that is, the more serious the network congestion. If the first device is a device that receives data packets, the time difference specifically represents the depth of queue congestion in the transmission direction from the second device to the first device. The smaller the time difference, the less congested the queue in the transmission direction from the second device to the first device; the larger the time difference, the less congested the queue in the transmission direction from the second device to the first device. The deeper the depth, the more serious the network congestion.

Exemplarily, adjusting the sending rate of the data message by the first device according to the time difference may include: if the time difference is less than a first preset threshold, increasing the sending rate of the data message; if the time difference is greater than the second preset threshold, reducing The sending rate of small data packets; the first preset threshold (T _low ) is less than the second preset threshold (T _high ), and the setting of the first preset threshold and the second preset threshold may be empirical values, and Link speed, device jitter and other factors are related.

For example, if the time difference T _{q is} less than T _low , it means that the depth of network queue congestion is very small, and the network can be considered not to be congested. In this case, the sending rate of data packets can be increased to fully utilize the network capacity; if T _{q is} greater than or equal to T _high , it means that the depth of network queue congestion is very large. It can be considered that the current network queue is relatively congested. This situation can reduce the sending rate of data packets to reduce the depth of network queue congestion; if T _{q is} greater than or equal to T _low , And is less than T _high , indicating that the depth of network queue congestion is within the acceptable range of the network, and the network can be considered lightly congested. In this case, the current data message sending rate may not be changed, and the light-congested network state can be maintained.

Exemplarily, the foregoing adjustment of the sending rate of the data message may increase or decrease the sending rate of the data message through a preset algorithm. For example, the algorithm can be a sum-increasing and multiplicative-decrease (AIMD) algorithm. The AIMD algorithm can be used to control the sending rate of data packets, including: when the network is not congested, linearly increasing its sending speed; When congested, it multiply reduces its sending speed. The embodiment of the present application does not limit the algorithm used to adjust the sending rate of the data message, and is only an exemplary description here.

It is understandable that the embodiment of the present application accurately obtains the time difference used to indicate the congestion depth of the network queue, and adjusts the sending rate of data packets according to the time difference, which can reduce the sending rate of data packets when the network is congested. , Thereby reducing the depth of network queue congestion and improving system performance.

In the network congestion control method provided by the embodiments of the present application, the first RTT and the second RTT are obtained, and the second RTT is subtracted from the first RTT to obtain the time difference; and according to the time difference, the sending rate of the data packet is adjusted to reduce Congestion depth of small network queues. In this embodiment, by using the third message and the fourth message with a priority higher than the priority of the data message, the second RTT can be measured more accurately, and the difference between the first RTT and the second RTT can be calculated to obtain the time difference. The time difference can accurately reflect the congestion depth of the network queue between the first device and the second device. Therefore, when adjusting the sending rate of data packets according to the time difference, the congestion depth of the network queue can be effectively reduced and system performance can be improved.

The embodiment of the present application also provides a network congestion control method. As shown in FIG. 7, the method further includes steps S701-S704 after step S307.

S701. If the first device determines that the first preset number of data packets have been sent cumulatively since the first message was sent last time, acquire the third RTT, or if the first device determines that the current time is different from the last time the first message was sent When the time interval reaches the first preset duration, the third RTT is obtained, and the current time stamp is recorded.

The third RTT and the first RTT are dynamic RTTs at different moments.

Exemplarily, since the degree of network congestion changes dynamically, the dynamic RTT can be detected periodically to determine the degree of current network congestion.

In an implementation manner, the cycle period of the cyclic detection of the dynamic RTT may be: the first device has cumulatively sent a first preset number of data packets since sending the first packet to obtain the third RTT. For example, since the last time the data packet carrying the first time stamp was sent, the first device has sent J data packets cumulatively, and J is greater than or equal to 2, and the first device obtains the third RTT.

In another implementation manner, the cycle period of the cyclic detection dynamic RTT may be: the first device determines that the time interval between the current time and the last sending of the first message reaches the first preset duration, obtains the third RTT, and records the current time stamp. For example, since the first message was sent last time, and the time interval reaches K microseconds, the first device obtains the third RTT.

The specific implementation manner of obtaining the third RTT in step S701 is the same as the specific implementation manner of obtaining the first RTT in the foregoing steps S301-S306. For details, reference may be made to the relevant description of the foregoing embodiment, and details are not described herein again.

S702. If the first device determines that the second preset number of data packets have been sent cumulatively since the last time the third message was sent, acquire the fourth RTT; or, if the first device determines that the current time is different from the last time the third message was sent When the time interval reaches the second preset duration, the fourth RTT is obtained, and the current time stamp is recorded.

The fourth RTT and the second RTT are fixed RTTs at different times.

Exemplarily, the data transmission path between the first device and the second device may change. For example, if there are multiple paths between the network card A of the computer node A and the network card B of the computer node B, when the first path between the network card A and the network card B is abnormal, the switch can choose the second path to reach the network card B for transmission data. It is understandable that after the network path between the network card A and the network card B changes, the fixed RTT will also change, so the fixed RTT can be detected periodically.

In an implementation manner, the cycle period of the cyclic detection of the fixed RTT may be: the first device has cumulatively sent a second preset number of data packets since sending the third message last time, and obtains the fourth RTT. For example, since the third message was sent last time, the first device has sent N data packets cumulatively, and N is greater than or equal to 2, and the first device obtains the fourth RTT.

In another implementation manner, the cycle period of cyclically detecting the fixed RTT may be: the first device determines that the time interval between the current time and the last sending of the third packet reaches the second preset duration, and obtains the fourth RTT. For example, since the third message was sent last time and the time interval reaches M microseconds, the first device obtains the fourth RTT.

The specific implementation manner of obtaining the fourth RTT in step S702 is the same as the implementation manner of obtaining the second RTT in the foregoing steps S601-S606. For details, reference may be made to the related description of the foregoing embodiment, and details are not repeated here.

Exemplarily, the cycle time for obtaining the third RTT in step S701 may be less than the cycle time for obtaining the fourth RTT in step S702, which is not limited in the embodiment of the present application, and is only an exemplary description here.

S703: Subtract the current fixed RTT from the third RTT to obtain the time difference.

Exemplarily, the current fixed RTT may be a fixed RTT saved in the context, and the fixed RTT saved in the context may be the second RTT or the fourth RTT.

(Optional) The first device may obtain the saved current fixed RTT from the context, and subtract the current fixed RTT saved in the context from the third RTT to obtain the time difference. If the current fixed RTT saved in the context is the second RTT, step S703 may subtract the second RTT from the third RTT to obtain the time difference. If the current fixed RTT saved in the context is the fourth RTT, step S703 may subtract the fourth RTT from the third RTT to obtain the time difference.

S704: Adjust the sending rate of the data message according to the time difference.

It can be understood that the specific implementation manner of step S704 may refer to the specific implementation manner in the foregoing step S307, which will not be repeated here.

It should be noted that the network congestion control method provided in this embodiment can repeatedly execute steps S701-S704 to control different network paths and different network congestion conditions at different times to ensure high network performance.

In this embodiment, the fixed RTT and the dynamic RTT are periodically and cyclically detected, which can improve the accuracy of the fixed RTT and the dynamic RTT, thereby effectively reducing the depth of network queue congestion and improving system performance.

FIG. 8 is a schematic diagram of a data transmission structure provided by an embodiment of the application. For example, if NIC A sends Write/Send to NIC B, NIC A is the device that sends data messages, and the transmission direction of the data message is from NIC A to NIC B, and the direction from NIC B to NIC A is no data transmission The direction of the message; if the network card A sends a Read to the network card B, then the network card A is the device that receives the data message, and the transmission direction of the data message is from the network card B to the network card A, and the direction from the network card A to the network card B is not transmitted The direction of the data message.

Take network card A sending a Write/Send request to network card B as an example, that is, network card A is the device that sends data packets. The network card A may include a fixed RTT request module 810, a fixed RTT response module 812, a dynamic RTT request module 820, a dynamic RTT response module 822, a rate control module 830, and a sending module 840. The network card B includes a fixed RTT reflection module 811, a dynamic RTT reflection module 821, and a receiving module 850.

The fixed RTT request module 810 is used to construct a fixed RTT request message and encapsulate the local time stamp 1 of the network card A in the message. The priority of the fixed RTT request message is higher than the priority of the data message sent by the network card A to the network card B.

The fixed RTT reflection module 811 is configured to receive the request message sent by the fixed RTT request module 810, and retrieve the time stamp 1 from the request message to construct a fixed RTT response message. The priority of the fixed RTT response message is higher than the priority of the data message sent by the network card A to the network card B.

The fixed RTT response module 812 is used to receive the fixed RTT response message sent by the fixed RTT reflection module 811, and extract the time stamp 1 from the fixed RTT response message, and take out the local time stamp 2 and take out according to the local time stamp 2 when the network card A receives the fixed RTT response message Time stamp 1, calculate the difference and save it in the context, and record the time difference as a fixed RTT.

The dynamic RTT request module 820 is configured to encapsulate the local time stamp 3 in the data message, and send the data message encapsulating the time stamp 3 to the network card B.

The receiving module 850 is configured to receive the data message sent by the dynamic RTT request module 820.

The dynamic RTT reflection module 821 is used to extract the time stamp 3 from the data message received by the receiving module 850 to construct a dynamic RTT response message. The priority of the dynamic RTT response message is higher than the priority of the data message.

The dynamic RTT response module 822 is used to receive the dynamic RTT response message sent by the dynamic RTT reflection module 821, and extract the time stamp 3 from the dynamic RTT response message, and take it out according to the local time stamp 4 when the network card A receives the dynamic RTT response message Time stamp 3, calculate the difference and save it in the context, and record the time difference as dynamic RTT.

The rate control module 830 is used to obtain the dynamic RTT from the dynamic RTT response module 822, and obtain the stored fixed RTT from the context, and calculate the time difference. The time difference (T _q ) is the dynamic RTT minus the fixed RTT, and the time difference is used to indicate the network Congestion level of the queue. If T _{q is} less than T _low , it means that the network queue is not congested and the data transmission rate can be increased; if T _{q is} greater than or equal to T _high , it means the network queue is congested, and the data transmission rate can be reduced to reduce the congestion depth of the queue; if T _{q is} greater than or equal to T _low and less than T _high , indicating that the network queue is lightly congested, and the current data sending rate is not changed.

The sending module 840 is a data transmission module for sending data messages. The sending module 840 may send data packets according to the sending rate adjusted by the rate control module 830.

It should be noted that FIG. 8 only takes the network card A as a device for sending data packets as an example for description. In actual applications, the network card B may also be a device for sending data, which is not limited in the embodiment of the present application. When the network card B is the end transmitting data, the functional modules included in the network card B are the same as the modules included in the network card A in FIG. 8.

The foregoing mainly introduces the solutions provided in the embodiments of the present application from the perspective of method steps. It can be understood that, in order to implement the above-mentioned functions, a computer includes hardware structures and/or software modules corresponding to each function. Those skilled in the art should easily realize that, in combination with the modules and algorithm steps of the examples described in the embodiments disclosed herein, this application can be implemented in a combination of hardware and computer software. Professionals and technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of this application.

The embodiment of the present application may divide the computer into functional modules according to the foregoing method examples. For example, each functional module may be divided corresponding to each function, or two or more functions may be integrated into one processing module. The above-mentioned integrated modules can be implemented in the form of hardware or software functional modules. It should be noted that the division of modules in the embodiments of the present application is illustrative, and is only a logical function division, and there may be other division methods in actual implementation.

In the case of dividing each function module corresponding to each function, FIG. 9 shows a possible structural schematic diagram of a network congestion control device involved in the above embodiment. The network congestion control device 900 includes: a processing module 901 and a transceiver module 902. The processing module 901 can execute S301, S305-S307 in FIG. 3, or S601, S605-S606 in FIG. 6, or S701-S704 in FIG. 7 through the transceiver module 902. Among them, all relevant content of the steps involved in the above method embodiments can be cited in the functional description of the corresponding functional module, which will not be repeated here.

In the case of dividing each functional module corresponding to each function, FIG. 10 shows a possible structural schematic diagram of a network congestion control device involved in the foregoing embodiment. The network congestion control device 1000 includes: a processing module 1001 and a transceiver module 1002. The processing module 1001 can execute S302-S304 in FIG. 3 or S602-S604 in FIG. 6 through the transceiver module 1002. Among them, all relevant content of the steps involved in the above method embodiments can be cited in the functional description of the corresponding functional module, which will not be repeated here.

In the case of using an integrated unit, FIG. 11 shows a schematic diagram of a possible structure of the network congestion control apparatus 1100 involved in the foregoing embodiment. The network congestion control device 1100 includes: a processor 1101 and a transceiver 1102. The processor 1101 is used to control and manage the actions of the network congestion control device 1100. For example, the processor 1101 is used to perform the operations shown in FIG. 3 through the transceiver 1102. S301, S305-S307, or S601, S605-S606 in FIG. 6, or S701-S704 in FIG. 7, and/or other processes used in the technology described herein. Optionally, the aforementioned network congestion control apparatus 1100 may further include a memory 1103 configured to store the program code and data corresponding to any of the network congestion control methods provided above by the network congestion control apparatus 1100. The memory 1103 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, random access memory (RAM), etc.

In the case of using an integrated unit, FIG. 12 shows a schematic diagram of a possible structure of the network congestion control apparatus 1200 involved in the foregoing embodiment. The network congestion control device 1200 includes a processor 1201 and a transceiver 1202. The processor 1201 is configured to control and manage the actions of the network congestion control device 1200. For example, the processor 1201 is configured to execute the operation shown in FIG. 3 through the transceiver 1202. S302-S304, or S602-S604 in FIG. 6, and/or other processes used in the techniques described herein. Optionally, the aforementioned network congestion control apparatus 1200 may further include a memory 1203 configured to store the program code and data corresponding to any of the network congestion control methods provided above by the network congestion control apparatus 1200. The memory 1203 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, random access memory (RAM), etc.

The steps of the method or algorithm described in conjunction with the disclosure of this application can be implemented in a hardware manner, or implemented in a manner in which a processor executes software instructions. Software instructions can be composed of corresponding software modules, which can be stored in random access memory (Random Access Memory, RAM), flash memory, erasable programmable read-only memory (Erasable Programmable ROM, EPROM), and electrically erasable Programming read-only memory (Electrically EPROM, EEPROM), register, hard disk, mobile hard disk, CD-ROM or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in the ASIC. In addition, the ASIC may be located in the core network interface device. Of course, the processor and the storage medium may also exist as discrete components in the core network interface device.

Those skilled in the art should be aware that in one or more of the above examples, the functions described in this application can be implemented by hardware, software, firmware or any combination thereof. When implemented by software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on the computer-readable medium. The computer-readable medium includes a computer storage medium and a communication medium, where the communication medium includes any medium that facilitates the transfer of a computer program from one place to another. The storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

The specific implementations described above further describe the purpose, technical solutions and beneficial effects of this application in detail. It should be understood that the above are only specific implementations of this application and are not intended to limit the scope of this application. The protection scope, any modification, equivalent replacement, improvement, etc. made on the basis of the technical solution of this application shall be included in the protection scope of this application.

Claims (22)

1. A network congestion control method, characterized in that it is applied to a first device, the first device being a device that sends data packets, and the method includes:

   The first device sends a first message to the second device, and the first message carries a first time stamp; the first time stamp is a local time stamp when the first message is sent;

   Receiving, by the first device, a second message sent by the second device, where the second message carries the first timestamp;

   Subtract the first time stamp from the second time stamp to obtain the first round-trip time RTT; the second time stamp is the local time stamp when the first device receives the second message;

   Adjusting the sending rate of data packets according to the first RTT;

   Wherein, the priority of the first message is the same as the priority of the data message, and the priority of the second message is higher than the priority of the data message.
2. The method of claim 1, wherein the method further comprises:

   Sending, by the first device, a third message to the second device, where the third message carries a third timestamp, and the third timestamp is a local timestamp when the third message is sent; The priority of the third message is higher than the priority of the data message;

   The first device receives the fourth message sent by the second device; the fourth message carries the third timestamp; the priority of the fourth message is higher than that of the data message priority;

   Subtract the third timestamp from the fourth timestamp to obtain the second RTT; the fourth timestamp is the local timestamp when the first device receives the fourth packet.
3. The method according to claim 2, wherein the adjusting the sending rate of the data packet according to the first RTT comprises:

   Subtract the second RTT from the first RTT to obtain a time difference; the time difference is used to indicate the depth of network queue congestion;

   According to the time difference, the sending rate of the data message is adjusted.
4. The method according to claim 3, wherein the adjusting the sending rate of the data message according to the time difference comprises:

   If the time difference is less than a first preset threshold, increase the sending rate of the data message; if the time difference is greater than a second preset threshold, reduce the sending rate of the data message; the first preset The threshold is less than the second preset threshold.
5. The method according to any one of claims 2-4, wherein the method further comprises:

   If the first device determines that the first preset number of data packets have been sent cumulatively since sending the first message last time, acquire the third RTT; or,

   If the first device determines that the time interval between the current time and the last sending of the first message reaches the first preset duration, obtain the third RTT, and record the current timestamp.
6. The method according to any one of claims 2-5, wherein the method further comprises:

   If the first device determines that the second preset number of data packets have been sent cumulatively since sending the third message last time, acquire the fourth RTT; or,

   If the first device determines that the time interval between the current time and the last sending of the third message reaches the second preset duration, obtains the fourth RTT, and records the current timestamp.
7. The method according to any one of claims 2-6, wherein the third timestamp is carried in a reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried in the message. In the payload of the text RDMA.
8. The method according to any one of claims 1-7, wherein if the first message is the data message, the first timestamp is carried in a reserved field in the BTH of the RDMA message If the first message and the data message are different, the first time stamp is carried in a reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried in the message RDMA The load Payload.
9. A network congestion control method, characterized in that the method includes:

   The second device receives the first message sent by the first device, and the first message carries a first time stamp; the first time stamp is the local time stamp when the first message is sent; One device is a device that sends data packets;

   Sending, by the second device, a second message to the first device, the second message carrying the first timestamp;

   Wherein, the priority of the first message is the same as the priority of the data message, and the priority of the second message is higher than the priority of the data message.
10. The method according to claim 9, wherein the method further comprises:

    The second device receives a third message sent by the first device, the third message carries a third timestamp, and the third timestamp is a local timestamp when the third message is sent ; The priority of the third message is higher than the priority of the data message;

    The second device sends a fourth message to the first device; the fourth message carries the third timestamp; the priority of the fourth message is higher than the priority of the data message level.
11. A network congestion control device, characterized in that the network congestion control device is a device for sending data messages, and the device includes: a processing unit and a transceiver unit;

    The processing unit is used to:

    Sending a first message to a second device through the transceiver unit, where the first message carries a first timestamp; the first timestamp is a local timestamp when the first message is sent;

    Receiving a second message sent by the second device through the transceiver unit, where the second message carries the first time stamp;

    Subtract the first time stamp from the second time stamp to obtain the first round trip time RTT; the second time stamp is the local time stamp when the device receives the second message;

    Adjusting the sending rate of data packets according to the first RTT;

    Wherein, the priority of the first message is the same as the priority of the data message, and the priority of the second message is higher than the priority of the data message.
12. The device according to claim 11, wherein the processing unit is further configured to:

    Sending a third message to the second device through the transceiving unit, where the third message carries a third timestamp, and the third timestamp is a local timestamp when the third message is sent; The priority of the third message is higher than the priority of the data message;

    The fourth message sent by the second device is received by the transceiver unit; the fourth message carries the third time stamp; the priority of the fourth message is higher than that of the data message priority;

    Subtract the third timestamp from the fourth timestamp to obtain the second RTT; the fourth timestamp is the local timestamp when the device receives the fourth message.
13. The device according to claim 12, wherein the processing unit is specifically configured to:

    Subtract the second RTT from the first RTT to obtain a time difference; the time difference is used to indicate the depth of network queue congestion;

    According to the time difference, the sending rate of the data message is adjusted.
14. The device according to claim 13, wherein the processing unit is specifically configured to:

    If the time difference is less than a first preset threshold, increase the sending rate of the data message; if the time difference is greater than a second preset threshold, reduce the sending rate of the data message; the first preset The threshold is less than the second preset threshold.
15. The device according to any one of claims 12-14, wherein the processing unit is further configured to:

    If the processing unit determines that the first preset number of data packets have been sent cumulatively since the first message was sent last time, acquire the third RTT; or,

    If the processing unit determines that the time interval between the current time and the last sending of the first message reaches the first preset duration, acquire the third RTT, and record the current time stamp.
16. The device according to any one of claims 12-15, wherein the processing unit is further configured to:

    If the processing unit determines that the second preset number of data packets have been sent cumulatively since sending the third message last time, acquire the fourth RTT; or,

    If the processing unit determines that the time interval between the current time and the last sending of the third message reaches the second preset duration, acquire the fourth RTT, and record the current time stamp.
17. The apparatus according to any one of claims 12-16, wherein the third time stamp is carried in a reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried in the message. In the payload of the text RDMA.
18. The apparatus according to any one of claims 11-17, wherein if the first message is the data message, the first timestamp is carried in a reserved field in the BTH of the message RDMA If the first message and the data message are different, the first time stamp is carried in a reserved field in the basic transmission header BTH of the remote direct memory access RDMA of the message, or carried in the message RDMA The load Payload.
19. A network congestion control device, characterized in that the device includes: a processing unit and a transceiver unit;

    The processing unit is used to:

    The first message sent by the first device is received by the transceiver unit, and the first message carries a first time stamp; the first time stamp is the local time stamp when the first message is sent; The first device is a device that sends data packets;

    Sending a second message to the first device through the transceiver unit, where the second message carries the first timestamp;

    Wherein, the priority of the first message is the same as the priority of the data message, and the priority of the second message is higher than the priority of the data message; or, the priority of the first message is The priority is higher than the priority of the data message, and the priority of the second message is the same as the priority of the data message.
20. The device according to claim 19, wherein the processing unit is further configured to:

    A third message sent by the first device is received through the transceiver unit, the third message carries a third time stamp, and the third time stamp is the local time stamp when the third message is sent ; The priority of the third message is higher than the priority of the data message;

    Send a fourth message to the first device through the transceiver unit; the fourth message carries the third time stamp; the priority of the fourth message is higher than the priority of the data message level.
21. A computer storage medium in which computer program code is stored, wherein when the computer program code runs on a processor, the processor is caused to execute any one of claims 1-10 The network congestion control method described in item.
22. A network congestion control device, characterized in that the network congestion control device includes:

    Transceiver, used to send and receive information, or to communicate with other network elements;

    Memory, used to store computer execution instructions;

    The processor is configured to execute the computer-executable instructions to implement the network congestion control method according to any one of claims 1-10.

Images