WO2021203985A1

WO2021203985A1 - Congestion information synchronizing method and related apparatus

Info

Publication number: WO2021203985A1
Application number: PCT/CN2021/083150
Authority: WO
Inventors: 林钦亮; 王巧灵
Original assignee: 华为技术有限公司
Priority date: 2020-04-09
Filing date: 2021-03-26
Publication date: 2021-10-14
Also published as: CN113518037A

Abstract

Disclosed in embodiments of the present application are a congestion information synchronizing method and a related apparatus, applied to a scenario of an in-network computing network, and used for filling in the gap of the lack of applicable congestion information synchronization in the in-network computing network and for causing the transmission rate of each of N working nodes to be smooth to avoid the interruption or even timeout of transmission caused by excessively fast transmission rates of the working nodes corresponding to communication links in which congestion does not occur. The congestion information synchronizing method comprises: obtaining congestion information, the congestion information being used for indicating that congestion occurs in a first communication link, the first communication link being a link between a first working node and an in-network computing switch, the first working node being any one of N working nodes, and N being an integer greater than 2; and sending to the N working nodes N packets having the same sequence number, the N packets having the same sequence number all carrying the congestion information, so that the N working nodes separately perform congestion control on the basis of the congestion information.

Description

Method and related device for synchronization of congestion information

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on April 9, 2020, the application number is 202010273713.0, and the invention title is "A method and related device for congestion information synchronization", the entire content of which is incorporated by reference In this application.

Technical field

The embodiments of the present application relate to the field of communication technologies, and in particular to a method and related devices for synchronizing congestion information.

Background technique

More and more network applications rely on large-scale computing, such as artificial intelligence, Internet of Things, cloud computing, and so on. To achieve large-scale computing, it is not feasible to rely on a single training node. Only through the collaborative computing of multiple training nodes in distributed computing can high-performance computing be provided for network applications. The in-net computing network makes full use of network computing resources, allocates some key calculations for distributed computing nodes, and provides aggregated calculations, so that multiple pieces of data are aggregated into one, thereby reducing network bandwidth usage and speeding up network transmission , Speed up distributed computing.

As we all know, congestion control is an important means to improve network resource utilization and optimize transmission quality. The quality of congestion handling directly affects the performance of the system. At present, standard remote direct memory access (RDMA) protocols use congestion control algorithms based on explicit congestion notification (ECN) flags, and more and more are based on transmission control protocols (transmission control protocols). The control protocol (TCP) protocol and applications have also begun to enable the ECN flag in the standard TCP protocol and the corresponding congestion control method. The congestion control method based on the ECN flag bit can be understood as: if the data flow is sent from node A to node B, when the switch in the network detects that the port is congested, it will pass the forward ECN position in all the packets of the port The bit is 1. If node B receives a message with forward ECN set to 1 (indicating that the route from node A to node B is congested), then the backward ECN bit in the message of node A will be set to 1. Or reply to a congestion notification packet (CNP), so that node A performs congestion control when it receives a packet with the ECN set or a CNP packet.

However, the current congestion control method based on the ECN flag is only suitable for point-to-point unicast communication. In the many-to-many synchronous communication mode calculated in the network, the current congestion control method based on the ECN flag is easy to work The data message sent by the node is discarded by the network computing switch in the process of aggregation processing; in addition, the communication mode becomes a many-to-many communication mode due to the process of providing aggregation calculation by the computing network within the network. However, the traditional point-to-point congestion control method is likely to cause the transmission rate of each working node in the many-to-many communication mode to be unable to synchronize, so that multiple working nodes cannot synchronize congestion control.

Summary of the invention

The embodiments of the present application provide a method and related devices for synchronization of congestion information, which are used to synchronously send congestion information to N working nodes, which fills in the gap of the lack of applicable congestion information synchronization in the intranet computing network, and further enables N The sending rate of each of the working nodes tends to be smooth.

In the first aspect, an embodiment of the present application provides a method for synchronizing congestion information, and the method may include:

The in-network computing switch acquires congestion information, where the congestion information is used to indicate that the first communication link is congested, and the first communication link is the link between the first working node and the in-network computing switch, and The first working node is any one of N working nodes, and the N is an integer greater than 2;

The computing switch in the network sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the congestion information, so that the N working nodes are based on The congestion information performs congestion control.

Through the above method, the computing switch in the network sends N packets with the same sequence number that all carry congestion information to N working nodes, so that the congestion information can be synchronized to these N working nodes, which fills in the lack of the computing network in the network. The applicable congestion information synchronization is blank, and the N working nodes can perform synchronous congestion control according to the congestion information carried in the message, so that the sending rate of each working node tends to be smooth.

Optionally, with reference to the above first aspect, in a first possible implementation manner, the in-network computing switch acquiring congestion information may include:

The in-network computing switch obtains the first message sent by the first working node, the first message carries the congestion information, and the congestion information includes the display congestion notification ECN flag bit of the first message The first value above, the first value is used to indicate that the first communication link is congested, and the first working node is any one of the N working nodes;

Correspondingly, the in-network computing switch sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the congestion information, including:

During the congestion indication aging period of the first value, the in-network computing switch sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the Congestion information.

Through the above method, when the congestion indication aging period of the first value has not expired, the network computing switch will carry the congestion information in N packets with the same sequence number, and send the N messages to N working nodes. A message with the same sequence number not only realizes that the congestion information can be synchronized to these N working nodes, and fills the gap of the lack of applicable congestion information synchronization in the intranet computing network, but also enables the N working nodes to be able to synchronize according to the report. The congestion information carried in the text performs synchronous congestion control, and further solves the failure problem of the congestion information in the in-network computing network based on the first value being within the time limit.

Optionally, in combination with the first possible implementation manner of the first aspect described above, in the second possible implementation manner, the in-network computing switch sends N reports with the same sequence number to the N working nodes. Before the article, you can also include:

The in-network computing switch modifies the value of the ECN flag bit of the second packet to the first value to obtain a third packet, and the second packet is the first aggregation within the validity period. For a completed message, the sequence number of the second message is the same as the sequence number of the third message;

Correspondingly, the in-network computing switch sending N packets with the same sequence number to the N working nodes includes:

The computing switch in the network sends N third messages to the N working nodes, the first value in each third message indicates that the corresponding working node performs congestion control, and the N The sequence numbers in the third message are the same.

In the above manner, the computing switch in the network realizes synchronization of congestion information by sending N third messages with the same sequence number to N working nodes, and further makes the first value in each third message indicate the corresponding The working nodes perform congestion control so that the sending rate of each working node tends to be smooth.

Optionally, in combination with the first to second possible implementation manners of the first aspect described above, in the third possible implementation manner, it may further include:

If the in-network computing switch receives the fourth packet sent by the first working node within the congestion indicator aging period of the first value, the value on the ECN flag bit in the fourth packet Is the first value;

The in-network computing switch ignores the first value in the fourth packet.

Through the above method, only the congestion information carried in the first message needs to be sent synchronously, and the congestion information in the fourth message is ignored, which not only avoids repeated transmission of congestion information, but also saves network resources.

Optionally, in combination with the second possible implementation manner of the first aspect described above, in a fourth possible implementation manner, the in-network computing switch modifies the value of the ECN flag bit of the second packet to the first A value, which can include:

When the ECN flag bit of the first packet includes the first ECN field, and the ECN flag bit of the second packet includes the second ECN field, the in-network computing switch will set the value in the second ECN field The value is modified to the first value in the first ECN field; or,

When the ECN flag bit of the first message includes the first forward display congestion notification FECN bit, and the ECN flag bit of the second message includes the second FECN bit, the in-network computing switch will The value in the second FECN bit is modified to the first value in the first FECN bit; or,

When the ECN flag bit of the first message includes the first backward display congestion notification BECN bit, and the ECN flag bit of the second message includes the second BECN bit, the in-network computing switch will The value in the second BECN bit is modified to the first value on the first BECN bit.

In the embodiment, since the value on the ECN flag bit has multiple forms in different protocols, the way that the calculation switch in the network modifies the value on the ECN flag bit of the second packet to the first value is also There can be multiple types, and through the above-mentioned modification, multiple application possibilities are provided for the subsequent synchronization of congestion information.

Optionally, in combination with the above-mentioned first aspect, in a fifth possible implementation manner, the in-network computing switch acquiring congestion information may include:

When the port status between the first working node and the in-network computing switch shows congestion, the in-network computing switch modifies the value of the ECN flag bit in the N data messages to be broadcast to Obtain congestion information, where the N data messages to be broadcast are messages with the same sequence number among the N working nodes;

Correspondingly, the in-network computing switch sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the congestion information, which may include:

The in-network computing switch sends the modified N data messages to be broadcast to the N working nodes, and the values on the ECN flag bits in the modified N data messages to be broadcast are used respectively Instruct the N working nodes to perform congestion control.

In an embodiment, the computing switch in the network can send these N modified data messages to be broadcast to N working nodes, so that each modified data message to be broadcast carries congestion information, so that the congestion information can be Synchronize to these N working nodes, filling the gap of the lack of applicable congestion information synchronization in the intranet computing network; and enabling the N working nodes to synchronize according to the congestion information carried in the modified data message to be broadcast Congestion control further smoothes the sending rate of each of the N working nodes.

Optionally, in combination with the fifth possible implementation manner of the first aspect described above, in the sixth possible implementation manner, the in-network computing switch calculates the value on the ECN flag bit in the N data packets to be broadcast Modifications include:

When the ECN flag in the data message to be broadcast includes the third ECN field, the in-network computing switch sets the value in the third ECN field; or,

When the ECN flag bit in the data message to be broadcast includes the third FECN bit, the in-network computing switch sets the value in the third FECN bit.

In the second aspect, the embodiments of the present application provide an in-network computing switch, and the in-network computing switch may include:

The acquiring unit is configured to acquire congestion information, where the congestion information is used to indicate that the first communication link is congested, and the first communication link is the link between the first working node and the computing switch in the network, so The first working node is any one of N working nodes, and the N is an integer greater than 2;

The sending unit is configured to send N messages with the same sequence number to the N working nodes, and the N messages with the same sequence number all carry the congestion information, so that the N working nodes are based on all the messages. The congestion information is described for congestion control.

Optionally, with reference to the above second aspect, in the first possible implementation manner, the acquiring unit may include:

The first obtaining module is configured to obtain a first message sent by the first working node, the first message carrying the congestion information, and the congestion information includes the display congestion notification ECN flag of the first message A first value on the bit, where the first value is used to indicate that the first communication link is congested;

Correspondingly, the sending unit includes:

The first sending module is configured to send N packets with the same sequence number to the N working nodes within the congestion indication time period of the first value obtained by the first obtaining module, and the N All packets with the same sequence number carry the congestion information.

Optionally, in combination with the first possible implementation manner of the second aspect described above, in the second possible implementation manner, the in-network computing switch may further include:

The modifying unit is configured to modify the value of the ECN flag bit of the second message to the first value before the first sending module sends N messages with the same sequence number to the N working nodes, To obtain a third message, where the second message is the first message to be aggregated within the validity period, and the sequence number of the second message is the same as the sequence number of the third message;

Correspondingly, the first sending module includes:

The first sending submodule is configured to send the N third messages obtained by the modifying unit to the N working nodes, and the first value in each third message indicates the corresponding work The node performs congestion control, and the sequence numbers in the N third messages are the same.

Optionally, in combination with the first to second possible implementation manners of the above second aspect, in a third possible implementation manner, the in-network computing switch further includes:

The acquiring unit is configured to receive a fourth message sent by the first working node within the congestion indication aging period of the first value, and the value on the ECN flag bit in the fourth message Is the first value;

The ignoring unit is configured to ignore the first value in the fourth message obtained by the obtaining unit.

Optionally, in combination with the second possible implementation manner of the second aspect described above, in the fourth possible implementation manner,

The modification unit is configured to, when the ECN flag bit of the first packet includes a first ECN field, and the ECN flag bit of the second packet includes a second ECN field, add the second ECN field to The value of is modified to the first value in the first ECN field; or,

The modification unit is configured to: when the ECN flag bit of the first message includes the first forward display congestion notification FECN bit, and the ECN flag bit of the second message includes the second FECN bit, then the The value in the second FECN bit is modified to the first value in the first FECN bit; or,

The modification unit is configured to: when the ECN flag bit of the first message includes the first backward display congestion notification BECN bit, and the ECN flag bit of the second message includes the second BECN bit, then the The value in the second BECN bit is modified to the first value on the first BECN bit.

Optionally, with reference to the above second aspect, in a fifth possible implementation manner, the acquiring unit may include:

The second acquisition module is configured to modify the value of the ECN flag bit in the N data packets to be broadcast when the port status between the first working node and the in-network computing switch shows congestion, To obtain congestion information, where the N data messages to be broadcast are messages with the same sequence number among the N working nodes;

Correspondingly, the sending unit includes:

The second sending module is configured to send the modified N data messages to be broadcast to the N working nodes, and the values on the ECN flag bits in the modified N data messages to be broadcast are respectively used Instruct the N working nodes to perform congestion control.

Optionally, in combination with the fifth possible implementation manner of the second aspect described above, in the sixth possible implementation manner, the second acquisition module is configured to include the ECN flag bit in the data message to be broadcast When the third ECN field is used, the calculation switch in the network sets the value in the third ECN field; or,

The second acquiring module is configured to, when the ECN flag bit in the data message to be broadcast includes the third FECN bit, the in-network computing switch sets the value in the third FECN bit.

In a third aspect, an embodiment of the present application provides a computer device, including: a processor and a memory; the memory is used to store program instructions, and when the computer device is running, the processor executes the program instructions stored in the memory to enable The computer device executes the congestion information synchronization method as described in the first aspect or any one of the possible implementation manners of the first aspect.

In a fourth aspect, embodiments of the present application provide a computer-readable storage medium, including instructions, which when run on a computer, cause the computer to execute a method as in the first aspect or any one of the possible implementation manners of the first aspect.

In the fifth aspect, the embodiments of the present application provide a computer program product containing instructions, which when run on a computer, cause the computer to execute the method as in the first aspect or any one of the possible implementation manners of the first aspect.

In a sixth aspect, an embodiment of the present application provides a chip system, which includes a processor, and is used to support the in-network computing switch to implement the functions involved in the first aspect or any one of the possible implementation manners of the first aspect. . In a possible design, the chip system also includes a memory, and the memory is used to store the program instructions and data necessary for calculating the switch in the network. The chip system can be composed of chips, or include chips and other discrete devices.

It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

In the embodiment of the present application, because the congestion information can indicate that the first communication link between any working node and the computing switch in the network is congested, when the computing switch in the network obtains the congestion information, the congestion information is all It is carried in N messages with the same sequence number, so that the N messages with the same sequence number are sent to N working nodes. In other words, the computing switch in the network sends N packets with the same sequence number that all carry congestion information to N working nodes, which not only realizes that the congestion information can be synchronized to these N working nodes, so that N working nodes After each message is received, the congestion control can be performed synchronously according to the congestion information carried in the message, which fills in the gap of the lack of applicable congestion information synchronization in the computing network in the network, and further makes the N working nodes The sending rate of each working node tends to be smooth.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings needed in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application.

FIG. 1 is a schematic diagram of a system architecture provided by an embodiment of the present application;

FIG. 2 is a schematic diagram of an embodiment of a method for synchronization of congestion information provided by this embodiment;

FIG. 3 is a schematic diagram of aggregation calculation performed by the in-network computing switch provided by this embodiment;

FIG. 4 is a schematic diagram of another embodiment of a method for synchronization of congestion information provided by this embodiment;

FIG. 5 is a schematic diagram of the state of the ECN flag bit of the RoCE v2 protocol or the TCP protocol proposed in the embodiment of the present application;

FIG. 6 is a schematic diagram of another embodiment of the method for synchronization of congestion information provided by this embodiment;

FIG. 7 is a schematic diagram of an embodiment of an in-network computing switch provided by an embodiment of the present application;

FIG. 8 is a schematic diagram of another embodiment of an in-network computing switch provided by an embodiment of the present application;

FIG. 9 is a schematic diagram of another embodiment of an in-network computing switch provided by an embodiment of the present application;

FIG. 10 is a schematic diagram of another embodiment of an in-network computing switch provided by an embodiment of the present application;

FIG. 11 is a schematic diagram of the hardware structure of a communication device in an embodiment of the present application.

Detailed ways

The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of this application and the above-mentioned drawings are used to distinguish similar objects, without having to use To describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations of them are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those clearly listed. Those steps or units may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or equipment.

Congestion control is an important means to improve network resource utilization and optimize transmission quality. The quality of congestion handling directly affects the performance of the system. Since the in-network computing network can make full use of network computing resources, it can allocate part of the key calculations for distributed computing nodes, and can provide aggregate calculations, so that multiple pieces of data can be aggregated into one, thereby reducing network bandwidth usage and speeding up network transmission. In-network computing networks that can provide aggregated computing are becoming more and more popular. And the intranet computing network has corresponding flow control features, namely: 1. The intranet computing switch performs aggregation calculation on packets with the same sequence number (index value) sent by N working nodes (N is an integer greater than 2), After calculating the parameters in the data message with the same serial number sent by N working nodes, it will be sent out, otherwise the packet will be lost; 2. In order to prevent the buffer area in the calculation switch in the network from overflowing, N working nodes are required to send data Synchronize the sending rate at the time; the sending rate of 3.N working nodes depends on the slowest link in the topology. The traditional congestion control method based on the ECN flag bit is only suitable for point-to-point unicast communication, and cannot be applied to the many-to-many synchronous communication mode of intranet computing network, that is, the congestion control method in point-to-point unicast communication is adopted. , It cannot solve that when any working node is congested and the sending rate is reduced, the sending rate of the other N-1 working nodes can also be reduced accordingly.

Therefore, in order to solve the above-mentioned problem, the method proposed in the embodiment of the present application is mainly applied to the application scenario of the in-network computing network that performs congestion control based on the ECN flag. The foregoing application scenarios of the in-network computing network include, but are not limited to, artificial intelligence (AI) distributed training, a mapping reduction model (MapReduce), or a distributed database, and so on. Based on the application scenario of the intranet computing network, please refer to FIG. 1. An embodiment of the present application provides a schematic diagram of a system architecture. It can be seen from Figure 1 that the system can include an in-network computing switch and N working nodes; among them, the in-network computing switch is mainly used to obtain congestion information, and because the congestion information can indicate any of the N working nodes The first communication link between a working node and the computing switch in the network is congested (the black dot in Figure 1). Therefore, when the first communication link is congested, the computing switch in the network can be in N All packets with the same sequence number carry the congestion information, so that the computing switch in the network can send N packets with the same sequence number to N working nodes. In this way, based on N packets with the same sequence number to carry the same congestion information, it not only realizes that the congestion information can be synchronized to the N working nodes, but also enables the N working nodes to receive the same N serial numbers. After the message is sent, congestion control can be performed synchronously according to the congestion information carried in the message, so as to avoid the situation that the working node corresponding to the communication link that is not congested causes transmission interruption or even timeout due to the excessively fast sending rate.

It should be understood that the first communication link that is congested shown in Figure 1 is the link between the computing switch in the network and the working node 0, which is only a schematic description. In actual applications, congestion occurs. The first communication link may also be a link between a working node such as a working node 1 or a working node 2 and a computing switch in the network, which will not be specifically limited in the embodiment of the present application.

It should be understood that, in addition to the traditional forwarding capabilities, the aforementioned intra-network computing switches also have certain programmability and computing capabilities, and can calculate and modify the fields in the message, such as modifying the ECN field and the The calculation result is substituted into the load of the message and so on. In addition, the aforementioned intra-network computing switches include but are not limited to Bareboot Wedge 100B switches, Cisco N3400 switches, etc., which will not be specifically limited in the embodiments of the present application. In addition, the aforementioned N working nodes may be servers with graphics processing units (GPUs), training nodes, etc., which will not be specifically limited in the embodiments of the present application.

The method for synchronizing congestion information in this embodiment is not only applicable to the system architecture shown in FIG. 1 above, but also applicable to other system architectures, which is not specifically limited here.

In order to facilitate a better understanding of the solution proposed in the embodiment of this application, the specific process in this embodiment will be introduced below. Please refer to FIG. 2, which is a schematic diagram of an embodiment of the method for synchronization of congestion information provided by this embodiment. Methods can include:

201. The computing switch in the network obtains congestion information, the congestion information is used to indicate that the first communication link is congested, the first communication link is the link between the first working node and the computing switch in the network, and the first working node is N Any one of the working nodes, N is an integer greater than 2.

In the embodiment, there is a corresponding communication link between each working node and the computing switch in the network, and the first communication link between any one of the N working nodes and the computing switch in the network In the event of congestion, the computing switches in the network will obtain congestion information, and then determine the occurrence of congestion based on the congestion information.

It is understandable that the congestion information can be that the computing switch in the network detects that any communication port leading to these N working nodes is in a congested state, and thus obtains the congestion information based on the communication port in the congested state; or The first working node corresponding to the first communication link where the congestion occurs notifies the congestion information to the in-network computing switch. It should be understood that the method for obtaining congestion information will not be specifically limited in the embodiment of the present application.

202. The computing switch in the network sends N packets with the same sequence number to N working nodes, and the N packets with the same sequence number all carry congestion information, so that the N working nodes respectively perform congestion control based on the congestion information.

In the embodiment, the serial number can indicate the serial number of the message, and for the messages with the same serial number, it indicates that the data messages sent by the N working nodes to the computing switch in the network belong to the same batch. The internal computing switch can distinguish the data packets sent by N working nodes based on the serial number, and correspond the parameters of the data packets from N different working nodes with the same serial number to the same batch of data packets. The aggregate calculation.

The foregoing aggregation calculation can be understood as that N working nodes synchronously send data packets carrying data to be calculated to the computing switches in the network, and different working nodes send different data packets that are numbered with serial numbers, and the computing switches in the network are After receiving N data packets, the parameters in the data packets with the same sequence number are aggregated accordingly. When all the data packets with a certain sequence number sent by N working nodes are calculated, the network The computing switch will send the aggregation result to N working nodes in the form of packets.

For example, please refer to Figure 3 for a schematic diagram of an in-network computing switch performing aggregation calculation. It can be seen from Figure 3 that the parameters in the data packets sent by the working node 0 to the computing switch in the network are "1", "2", and "3" respectively; The parameters in the data message are "4", "5", and "6"; the parameters in the data message sent by the working node 2 to the computing switch in the network are "7", "8", and "9" respectively. . The serial numbers of the data messages corresponding to the parameters "1", "4", and "7" (assuming index0) are the same, while the data corresponding to the parameters "2", "5", and "8" The serial numbers of the messages (assuming index1) are the same, and the serial numbers of the data messages corresponding to the parameters "3", "6", and "9" (assuming index2) are the same.

Therefore, the computing switch in the network will sum and average the parameters in the data packets with the same sequence number sent from working node 0, working node 1, and working node 2, for example, the parameters in the data packet corresponding to index 0 "1", "4", and "7" are averaged, and the polymerization result is 4. In this way, the calculation switch in the network will only send the aggregation result corresponding to the sequence number to the corresponding working node after calculating the parameters in the data message with the same sequence number sent by the three working nodes. It is understandable that the working node 0, working node 1 and working node 2 in Fig. 3 are only a schematic description for the aggregation calculation. In practical applications, the number of working nodes involved in the aggregation calculation The number is not limited, as long as N is an integer greater than 2.

Therefore, in order to synchronize the congestion information to N working nodes, the network computing switch needs to carry the congestion information in N packets with the same sequence number, so that these N packets with the same sequence number can be transferred. Send to N working nodes. That is to say, the computing switch in the network needs to carry the congestion information in the message obtained after the aggregation is completed, and send the message that carries the congestion information after the aggregation is completed to N working nodes, because only to N working nodes All messages with the same sequence number can be sent to the N working nodes to obtain the congestion information in the same batch of messages, which means that the congestion information sent to the N working nodes is synchronized, so that These N working nodes can perform synchronous congestion control based on the congestion information in the received message, avoiding the situation that the working node corresponding to the communication link that is not congested causes the transmission to be interrupted or even timed out due to the excessively fast transmission rate. .

In addition, it should be understood that the N working nodes respectively perform congestion control based on the congestion information, which can be understood as the N working nodes synchronously reducing the rate of sending messages corresponding to each other, etc., which will not be specifically done in this embodiment of the application. Limited description.

Based on the embodiment corresponding to Figure 2 above, it can be seen that the in-network computing switch can obtain congestion information in a variety of ways, and realize the synchronization of congestion information in different ways, which will be described in detail in the following embodiments:

Case 1: The working node corresponding to the congested communication link informs the computing switch in the network.

Case 2: Active detection by the computing switch in the network.

1. For the above situation 1, please refer to FIG. 4, which is a schematic diagram of another embodiment of the method for synchronizing congestion information provided by the embodiment of this application. As shown in FIG. 4, another embodiment of the method for synchronizing congestion information provided by the embodiment of the present application may include:

401. The in-network computing switch obtains the first message sent by the first working node, the first message carries congestion information, and the congestion information includes the first value on the ECN flag bit of the display congestion notification of the first message, and the first value is used To indicate that the first communication link is congested.

In the embodiment, the communication port of the computing switch in the network is not congested, but it is congested with any communication port of other switches connected to the computing switch in the network. The first working node here can be understood as passing the first working node. A communication link is indirectly connected to the computing switch in the network, then the first working node corresponding to the first communication link that is congested will carry the congestion information in the first message and send it to the computing switch in the network. .

It is understandable that the congestion information may include the first value on the ECN flag bit of the first packet, and the ECN flag bit in the first packet is a field with a length of 2 bits located in the header of the packet. For example: in remote direct memory access (remote direct memory access overconverged ethernet, RDMA overconverged ethernet, RoCE) v2 protocol or TCP protocol based on converged Ethernet, the aforementioned ECN flag is located in the fourth version of the Internet communication protocol (Internet protocol version 4, IPv4) or Internet protocol version 6 (Internet protocol version 6, IPv6) in the 2bit field of the packet header; while in the wireless broadband technology (InfiniBand, IB) protocol, the ECN flag is located in the basic transmission The 2-bit field of the header (base transport header, BTH) can be specifically composed of forward explicit congestion notification (FECN) or backward explicit congestion notification (BECN), that is, the first The one bit is the FECN bit, and the latter bit is the BECN bit, which will not be specifically limited in the embodiment of the present application.

In addition, the first value is the value on the ECN flag bit of the aforementioned first message, which can be used to indicate that the first communication link is congested. For example, please refer to FIG. 5, which is a schematic diagram of the state of the ECN flag bit of the RoCE v2 protocol or the TCP protocol proposed in the embodiment of this application. It can be seen from Figure 5 that when the value of the ECN field in the IPv4 or IPv6 packet header is 11, the corresponding state is the forward congestion flag, indicating that congestion occurs. Therefore, the ECN in the RoCE v2 protocol or the TCP protocol The first value on the flag bit may take a value of 11, which is used to indicate that the first communication link between the first working node and the computing switch in the network is congested.

In the ECN flag bit in the InfiniBand protocol, the value of the FECN bit is 1 indicating that congestion occurs, and the value of the BECN bit is 1 also indicating congestion. At this time, the value of the FECN bit is 1, which means that A is experiencing congestion in the process of sending data stream to B; if the data stream flows from the working node B to the working node A, and the value of the BECN bit is 1, it means that B is sending to A Congestion is encountered in the process of data flow. Therefore, the first value on the ECN flag bit in the InfiniBand protocol may include: the value of the first bit is 1, or the value of the second bit is 1.

Therefore, in the RoCE v2 protocol or the TCP protocol, the congestion information can be expressed as IP.ECN=11, and in the InfiniBand protocol, the congestion information can be expressed as InfiniBand.FECN=1 or InfiniBand.BECN=1. However, it should be understood that, in addition to the aforementioned definitions, the value of the ECN field, the FECN bit, or the BECN bit may also use other values to indicate the occurrence of congestion in practical applications, which will not be specifically limited in the embodiment of the present application.

402. During the congestion indication aging period of the first value, the in-network computing switch sends N packets with the same sequence number to N working nodes, and the N packets with the same sequence number all carry congestion information.

On the basis of the aggregation calculation described in Figure 3, assuming that the parameters "4" and "7" in the data packets corresponding to index0 sent by working node 1 and working node 2 have reached the calculation switch in the network, then at this time The calculation switch in the network will always wait for the working node 0 to send the parameter "1" in the data message corresponding to index0. After receiving "1", the calculation switch in the network will further target all the data corresponding to index0. Packets are aggregated and calculated. However, if the working node 0 is congested on the first communication link, the calculation switch in the network cannot receive the corresponding data packets, so that the aggregation calculation cannot be performed, and the buffer area of the calculation switch in the network is limited. Working node 1 and working node 2 that are not in congestion will continuously send data packets to the computing switch in the network, which can easily cause the buffer area of the computing switch in the network to overflow and be exhausted. Based on the calculation in the network, the transmission rate in the network often depends on the flow control characteristics of the slowest link in the topology. Before the aggregation of a data message with a certain sequence number is completed, the data message sent by the working node will be cached. Up or being discarded, causing the congestion information to become outdated and invalid.

Therefore, under the characteristics of strong timeliness of congestion information, when the in-network computing switch detects that the first message obtained from the first working node carries congestion information, it will pass a timer, a message timer, etc. Monitor the aging period of the congestion information in the first message.

When the congestion indication aging period of the first value has not expired, the computing switch in the network will carry the congestion information in the N packets with the same sequence number, and send the same N sequence numbers to the N working nodes. The message not only realizes that the congestion information can be synchronized to these N working nodes, so that after receiving these N messages with the same sequence number, the N working nodes can synchronize according to the congestion information carried in the message Local congestion control fills the gap in the lack of applicable congestion information synchronization in the intranet computing network, and can also solve the problem of failure of congestion information in the intranet computing network based on the first value being within the time limit.

Optionally, in some embodiments, before the in-network computing switch sends N packets with the same sequence number to N working nodes, the method for synchronizing congestion information may further include:

The calculation switch in the network modifies the value of the ECN flag bit of the second packet to the first value to obtain the third packet. The second packet is the first packet to be aggregated within the validity period. The serial number of the message is the same as the serial number of the third message;

Correspondingly, the computing switch in the network sends N packets with the same sequence number to N working nodes, including:

The computing switch in the network sends N third messages to N working nodes, and the first value in each third message indicates that the corresponding working node performs congestion control, and the sequence numbers in the N third messages are the same.

That is to say, after obtaining the first message, the in-network computing switch first parses the first message, and based on the flow control feature that the in-network computing switch will only send the message out after completing the aggregation calculation. Therefore, regardless of whether the first packet obtained after parsing is a packet of the first type or a packet of the second type, the computing switch in the network needs to wait for the first aggregation within the congestion indicator aging period of the first value. The completed message, that is, the second message, and then modify the value of the ECN flag bit of the second message to the first value. The purpose is to carry the congestion information in the second message, so that it can be The modified second message is taken as the third message carrying congestion information.

In this way, the computing switch in the network can send N third messages to N working nodes to synchronize the congestion information, so that the first value in each third message indicates that the corresponding working node performs congestion control, and further The ground makes the sending rate of each working node tend to be smooth.

It should be noted that, distinguishing whether the first message belongs to the first type of message or the second type of message can be determined by the message length of the first message. For example: when the message length of the first message is within the first preset message length, it can be determined that the first message belongs to the first type of message; when the message length of the first message is in the first 2. When the message length is within the preset message length, it can be determined that the first message belongs to the message of the second type. In addition, the aforementioned first preset message length is greater than the second preset message length, and the first type of message can be understood to be able to perform aggregation calculations, or directly broadcast data messages; and the second type of messages It can be understood that neither aggregate calculations nor broadcast data packets can be performed.

Optionally, in other embodiments, since the value on the ECN flag bit can take multiple forms in different protocols, the in-network computing switch modifies the value on the ECN flag bit of the second packet to There can also be multiple ways of the first value, which can be understood with reference to the following ways:

Method 1: When the ECN flag bit of the first packet includes the first ECN field and the ECN flag bit of the second packet includes the second ECN field, the in-network computing switch will modify the value in the second ECN field to the first The first value in the ECN field.

That is to say, in the RoCE v2 protocol or the TCP protocol, if the ECN flag bit in the first message received is the first ECN field, then the value in the second ECN field in the second message is modified to Same as the first value, such as: modified to "11", so that the congestion information carried in the first message can be copied to the second message, which provides multiple application possibilities for subsequent synchronization of congestion information sex.

It should be understood that setting the value of the first ECN field to "11" to indicate that congestion occurs is only a schematic description. In practical applications, it is also possible to define the value of the first ECN field as Other numerical values indicate the occurrence of congestion, which will not be specifically limited in the embodiment of the present application.

Manner 2: When the ECN flag bit of the first message includes the first forward display congestion notification FECN bit, and the ECN flag bit of the second message includes the second FECN bit, the intra-network computing switch sets the FECN bit in the second FECN bit The value is modified to the first value in the first FECN bit; or,

Method 3: When the ECN flag bit of the first message includes the first backward display congestion notification BECN bit, and the ECN flag bit of the second message includes the second BECN bit, the on-net computing switch will set the value in the second BECN bit The value is modified to the first value on the first BECN bit.

In the embodiment, for the second and third modes, the ECN flag in the InfiniBand protocol is located in the 2bit field in the BTH header, and the 2bit field is composed of FECN and BECN, and the value on the FECN bit is "1", or the value "1" on the BECN bit can indicate congestion.

Therefore, in the InfiniBand protocol, if the ECN flag bit in the first message received is the first FECN bit, then the value in the second FECN bit in the second message is modified to be the same as the first value. Such as: modify to "1". Or, if the ECN flag bit in the first message received is the first BECN bit, then the value in the second BECN bit in the second message is modified to be the same as the first value, such as: 1", so that the congestion information carried in the first message can be copied to the second message, which provides multiple application possibilities for subsequent synchronization of congestion information.

It should be understood that the value of the second FECN bit is set to "1" or the value of the second BECN bit is set to "1" to indicate that congestion occurs. This is only a schematic description, and in practical applications It is also possible to define the value on the second FECN bit or the value on the second BECN bit as other numerical values to indicate the occurrence of congestion, which will not be specifically limited in the embodiment of this application.

In addition, in order to save network resources and avoid repeated transmission of congestion information, in other embodiments, the method of congestion information synchronization may further include:

If the in-network computing switch receives the fourth message sent by the first working node during the congestion indication aging period of the first value, the value on the ECN flag bit in the fourth message is the first value;

The calculation switch in the network ignores the first value in the fourth packet.

In the embodiment, if within the congestion indication aging period of the first value, the network computing switch receives the fourth message sent by the first working node, at this time, because the value on the ECN flag bit in the fourth message is the same as that of the first working node. The first value in the congestion information carried in one message is the same, it means that the fourth message also carries the congestion information. However, the network computing switch has started a timer to monitor the aging of the first value when it receives the first message, and in order to avoid repeated copying and sending of congestion information, if the congestion indication is at the first value at this time The fourth packet that also carries congestion information is also received within the validity period. The calculation switch in the network does not need to restart another timer when the fourth packet is received, but ignores the fourth packet. The first value. That is to say, the computing switch in the network can ignore the congestion information in the fourth packet, and forward the fourth packet to the destination according to the original forwarding rules, that is, only need to perform the congestion information carried in the first packet. Just send it synchronously, saving network resources.

In addition, it can also be understood that if the congestion of the first value indicates that the aging is timed out, the in-network computing switch ignores the congestion information. In other words, the congestion indication aging at the first value has expired, then the congestion information corresponding to the first value has expired. At this time, the network computing switch will synchronize the invalid congestion information and it will not be able to make N Each working node performs congestion control synchronously, so the computing switch in the network can ignore the congestion information in the first message, send the first message to the destination according to the original forwarding rules, and re-transmit the congested communication link Obtain other congestion information that has not failed from the corresponding working node.

2. In view of the second situation described above, please refer to FIG. 6, which is a schematic diagram of another embodiment of a method for synchronization of congestion information provided in an embodiment of this application. As shown in FIG. 6, another embodiment of the method for synchronizing congestion information provided by the embodiment of the present application may include:

601. When the port status between the first working node and the computing switch in the network shows congestion, the computing switch in the network modifies the value of the ECN flag in the N data packets to be broadcast to obtain congestion information. Among them, the N data messages to be broadcast are messages with the same sequence number among the N working nodes.

In an embodiment, the in-network computing switch can monitor the cache queue information where the messages sent by N working nodes are located. When the cache queue information has exceeded the cache threshold, the in-network computing switch can determine among the N working nodes. The port status between any of the working nodes and the computing switch in the network has shown congestion, that is, the port status between the first working node and the computing switch in the network shows congestion.

It should be noted that the first working node here can be understood as a working node that is directly connected to the computing switch in the network through the first communication link. In addition, the first working node is only any one of the N working nodes. It is not limited in the embodiment of this application; secondly, in addition to the way that the in-network computing switch judges whether the port is congested by caching queue information, in practical applications, it can also determine the port by other judgment methods such as port utilization. It is in a congested state, which will not be specifically limited in the embodiment of the present application.

Therefore, when the port is congested, the computing switch in the network can modify the value of the ECN flag in the N data packets to be broadcast to obtain congestion information, for example, modify it to "ECN=1" or "ECN =11" and so on, then the corresponding congestion information can be expressed as "InfiniBand.ECN=1" or "IP.ECN=11" and so on.

In addition, it should be noted that the aforementioned N data messages to be broadcast are messages with the same sequence number obtained after the calculation switch in the network completes the aggregation calculation of the data messages with the same sequence number sent by the N working nodes. . The description of the serial number can be understood with reference to step 202 in FIG. 2, and details will not be repeated here.

602. The computing switch in the network sends the modified N data messages to be broadcast to N working nodes, and the values on the ECN flag bits in the modified N data messages to be broadcast are used to instruct the N working nodes to perform Congestion control.

In the embodiment, after modifying the value of the ECN flag bit in the N data messages to be broadcast, the obtained N modified data messages to be broadcast all carry congestion information, so the network calculation switch Then the N modified data messages to be broadcast can be sent to N working nodes, which not only realizes that the congestion information can be synchronized to these N working nodes, but also fills the gap of the lack of applicable congestion information synchronization in the intranet computing network. ; And so that after receiving the modified data message to be broadcast, the N working nodes can perform synchronous congestion control according to the congestion information carried in the modified data message to be broadcast, and further make the N working nodes The sending rate of each working node in the network tends to be smooth, so as to avoid the situation that the working node corresponding to the communication link that is not congested causes transmission interruption or even timeout because the sending rate is too fast.

Optionally, in other embodiments, since the value on the ECN flag bit can take multiple forms in different protocols, the in-network computing switch modifies the value on the ECN flag bit of the second packet to There can also be multiple ways of the first value. Therefore, the calculation switch in the network can modify the value of the ECN flag in the N data messages to be broadcast, which can be understood by referring to the following ways:

Manner 1: When the ECN flag in the data message to be broadcast includes the third ECN field, the in-network computing switch sets the value in the third ECN field.

That is to say, in the RoCE v2 protocol or TCP protocol, if the ECN flag bit in each data message to be broadcast is the third ECN field, then the third ECN field in each data message to be broadcast The value of is set, such as "11", so that the congestion information "IP.ECN=11" can be copied to N data messages to be broadcast, which provides multiple application possibilities for subsequent synchronization of congestion information .

It should be understood that setting the value of the third ECN field to "11" indicates that congestion occurs, which is only a schematic description. In practical applications, it is also possible to set the value of the third ECN field. Other values are used to indicate the occurrence of congestion, which will not be specifically limited in the embodiment of the present application.

Or, the second method: when the ECN flag bit in the data message to be broadcast includes the third FECN bit, the in-network computing switch sets the value in the third FECN bit.

In the embodiment, because the ECN flag bit in the InfiniBand protocol is located in the 2bit field in the BTH header, and the 2bit field is composed of the FECN bit and the BECN bit, and the value of the FECN bit is "1", or BECN The value "1" in the bit can indicate that congestion occurs. However, in the case of active detection by the computing switch in the network, only the value of the FECN bit needs to be modified to indicate the occurrence of congestion. Therefore, in the InfiniBand protocol, if the ECN flag bit in each data message to be broadcast is the third FECN bit, then the value in the third FECN bit in each data message to be broadcast is set. For example, "1", so that the congestion information "InfiniBand.FECN=1" can be copied into N data messages to be broadcast, which provides multiple application possibilities for subsequent synchronization of congestion information.

It should be understood that setting the value of the third FECN bit to "1" to indicate that congestion occurs is only a schematic description. In practical applications, it is also possible to set the value of the third FECN bit. Other values are used to indicate the occurrence of congestion, which will not be specifically limited in the embodiment of the present application.

In the embodiment of this application, the computing switch in the network sends N packets with the same sequence number that all carry congestion information to N working nodes, which not only realizes that the congestion information can be synchronized to these N working nodes, so that N After receiving the message, the working nodes can perform synchronous congestion control according to the congestion information carried in the message, filling in the gap of the lack of applicable congestion information synchronization in the computing network in the network, and further making the N working nodes The sending rate of each working node tends to be smooth.

The foregoing mainly introduces the congestion information synchronization method provided by the embodiment of the present application from the perspective of the method. It can be understood that in order to realize the above-mentioned functions, corresponding hardware structures and/or software modules for performing each function are included. 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 in the present application, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed by hardware or computer software-driven hardware depends on the specific application and design constraint conditions of the technical solution. 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 embodiments of the present application may divide the device 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.

The in-network computing switch in the embodiment of the present application will be described in detail below. Referring to FIG. 7, an embodiment of the in-network computing switch in the embodiment of the present application includes:

The obtaining unit 701 is configured to obtain congestion information, and the congestion information is used to indicate that a first communication link is congested, and the first communication link is a link between a first working node and the computing switch in the network, and The first working node is any one of N working nodes, and N is an integer greater than 2;

The sending unit 702 is configured to send N messages with the same sequence number to N working nodes. The N messages with the same sequence number all carry the congestion information obtained by the obtaining unit 701, so that the N working nodes are based on congestion. Information is used for congestion control.

In the above manner, the sending unit 702 sends to N working nodes N packets with the same sequence numbers that all carry the congestion information obtained by the obtaining unit 701, so that the congestion information can be synchronized to the N working nodes, filling the network. In the internal computing network, there is a lack of applicable congestion information synchronization gaps, and N working nodes can synchronize congestion control according to the congestion information carried in the message, so that the sending rate of each working node tends to be smooth.

For ease of understanding, on the basis of the embodiment described in FIG. 7, please refer to FIG. 8. In another embodiment of the in-network computing switch in the embodiment of the present application, the obtaining unit 701 may include:

The first obtaining module 7011 is configured to obtain the first message sent by the first working node, the first message carries congestion information, and the congestion information includes the first value on the ECN flag bit of the display congestion notification of the first message. The value is used to indicate that the first communication link is congested, where the first working node is any one of the N working nodes;

Correspondingly, the sending unit 702 may include:

The first sending module 7021 is configured to send N packets with the same sequence number and N packets with the same sequence number to N working nodes within the congestion indication aging period of the first value obtained by the first acquisition module 7011 Both carry congestion information.

In the above manner, when the congestion indication aging period of the first value has not expired, the first sending module 7021 carries the congestion information obtained by the first obtaining module 7011 in all N packets with the same sequence number, and Sending these N messages with the same sequence number to N working nodes, not only realizes that congestion information can be synchronized to these N working nodes, and fills the gap in the lack of applicable congestion information synchronization in the intranet computing network, but also It can enable N working nodes to perform synchronous congestion control according to the congestion information carried in the message, and further solve the failure problem of the congestion information in the in-network computing network based on the first value being within the validity period.

Optionally, on the basis of the embodiment described in FIG. 8, please refer to FIG. 9. In another embodiment of the in-network computing switch in the embodiment of the present application, the in-network computing switch may further include:

The modifying unit 703 is configured to modify the value of the ECN flag bit of the second message to the first value before the first sending module 7021 sends N messages with the same sequence number to the N working nodes to obtain the third Message, the second message is the first message to be aggregated within the validity period, and the sequence number of the second message is the same as the sequence number of the third message;

Correspondingly, the first sending module 7021 includes:

The first sending sub-module 70211 is configured to send N third messages obtained by the modification unit 703 to N working nodes. The first value in each third message indicates that the corresponding working node performs congestion control. The sequence numbers in the three messages are the same.

Optionally, on the basis of the embodiments described in FIG. 8 and FIG. 9, in another embodiment of the in-network computing switch in the embodiment of the present application, the in-network computing switch further includes:

The acquiring unit 701 is configured to receive a fourth message sent by the first working node within the congestion indication aging period of the first value, and the value on the ECN flag bit in the fourth message is the first value;

The ignoring unit is used to ignore the first value in the fourth message obtained by the obtaining unit 701.

Optionally, on the basis of the embodiment described in FIG. 9 above, in another embodiment of the in-network computing switch in the embodiment of the present application, the modifying unit 703 is configured to include the ECN flag bit of the first packet When the first ECN field and the ECN flag bit of the second packet include the second ECN field, modify the value in the second ECN field to the first value in the first ECN field; or,

The modification unit 703 is configured to: when the ECN flag bit of the first message includes the first forward display congestion notification FECN bit, and the ECN flag bit of the second message includes the second FECN bit, change the value in the second FECN bit Modify to the first value in the first FECN bit; or,

The modifying unit 703 is configured to: when the ECN flag bit of the first message includes the first backward display congestion notification BECN bit, and the ECN flag bit of the second message includes the second BECN bit, the value in the second BECN bit is changed Modified to the first value on the first BECN bit.

Optionally, on the basis of the embodiment described in FIG. 7, please refer to FIG. 10. In another embodiment of the in-network computing switch in the embodiment of the present application, the obtaining unit 701 may include:

The second acquisition module 7012 is used to modify the value of the ECN flag bit in the N data packets to be broadcast when the port status between the first working node and the in-network computing switch shows congestion to obtain the congestion Information, where the N data messages to be broadcast are messages with the same sequence number among the N working nodes, and the second working node is any one of the N working nodes;

Correspondingly, the sending unit 702 includes:

The second sending module 7022 is used to send modified N data messages to be broadcast to N working nodes, and the values on the ECN flag bits in the modified N data messages to be broadcast are used to indicate N jobs respectively The node performs congestion control.

In an embodiment, the second sending module 7022 may send the N modified data messages to be broadcast to N working nodes, so that each modified data message to be broadcast carries the data obtained by the second obtaining module 7012 Congestion information, so that the congestion information can be synchronized to these N working nodes, filling the gap of the lack of applicable congestion information synchronization in the computing network in the network; and allowing the N working nodes to be based on the modified data message to be broadcast The congestion information carried in the congestion control is performed synchronously, which further makes the sending rate of each of the N working nodes tend to be smooth.

Optionally, on the basis of the embodiment described in FIG. 10, in another embodiment of the in-network computing switch in the embodiment of the present application, the second obtaining module 7012 is used for the ECN flag in the data message to be broadcast When the bit includes the third ECN field, the calculation switch in the network sets the value in the third ECN field; or,

The second acquisition module 7012 is configured to set the value in the third FECN bit by the in-network computing switch when the ECN flag bit in the data message to be broadcast includes the third FECN bit.

The foregoing describes the intra-network computing switch in the embodiment of the present application from the perspective of a modular functional entity, and the following describes the intra-network computing switch in the embodiment of the present application from the perspective of hardware processing. FIG. 11 is a schematic diagram of the hardware structure of a communication device in an embodiment of the present application. As shown in FIG. 11, the communication device may include:

The communication device includes at least one processor 1101, a communication line 1107, a memory 1103, and at least one communication interface 1104.

The processor 1101 may be a general-purpose central processing unit (central processing unit, CPU), a microprocessor, an application-specific integrated circuit (server IC), or one or more programs for controlling the execution of the program of this application Integrated circuits.

The communication line 1107 may include a path to transmit information between the aforementioned components.

Communication interface 1104, which uses any device such as a transceiver to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc. .

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), or other types that can store information and instructions The dynamic storage device, the memory can exist independently, and is connected to the processor through the communication line 1107. The memory can also be integrated with the processor.

The memory 1103 is used to store computer-executed instructions for executing the solution of the present application, and the processor 1101 controls the execution. The processor 1101 is configured to execute computer-executable instructions stored in the memory 1103, so as to implement the congestion information synchronization method provided in the foregoing embodiment of the present application.

Optionally, the computer-executable instructions in the embodiments of the present application may also be referred to as application program codes, which are not specifically limited in the embodiments of the present application.

In a specific implementation, as an embodiment, the communication device may include multiple processors, such as the processor 1101 and the processor 1102 in FIG. 11. Each of these processors can be a single-CPU (single-CPU) processor or a multi-core (multi-CPU) processor. The processor here may refer to one or more devices, circuits, and/or processing cores for processing data (for example, computer program instructions).

In a specific implementation, as an embodiment, the communication apparatus may further include an output device 1105 and an input device 1106. The output device 1105 communicates with the processor 1101 and can display information in a variety of ways. The input device 1106 communicates with the processor 1101 and can receive user input in a variety of ways. For example, the input device 1106 may be a mouse, a touch screen device, a sensor device, or the like.

The aforementioned communication device may be a general-purpose device or a dedicated device. In a specific implementation, the communication device may be a router, an in-network computing switch, or a device with a similar structure in FIG. 11. The embodiment of the present application does not limit the type of the communication device.

The above-mentioned acquisition unit 701, first acquisition module 7011, and second acquisition module 7012 can all be implemented by the input device 1106, and the sending unit 702, the first sending module 7021, the first sending sub-module 70211, and the second sending module 7022 can all be implemented through the input device 1106. The output device 1105 is implemented, and both the modification unit 703 and the ignoring unit can be implemented by the processor 1101 or the processor 1102.

In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, it can be implemented in the form of a computer program product in whole or in part.

Those skilled in the art can clearly understand that, for the convenience and conciseness of the description, the specific working process of the above-described system, device, and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed system, device, and method can be implemented in other ways. For example, the device embodiment described above is only illustrative. For example, the division of the unit is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components may be combined or may be Integrate into another system, or some features can be ignored or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.

The unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may also be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

In addition, the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or software functional unit.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solution of the present application essentially or the part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , Including several instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method in each embodiment of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disks or optical disks and other media that can store program codes. .

The above embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still compare the previous embodiments. The recorded technical solutions are modified, or some of the technical features are equivalently replaced; these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

A method for synchronizing congestion information is characterized in that it includes:

The in-network computing switch acquires congestion information, where the congestion information is used to indicate that the first communication link is congested, and the first communication link is the link between the first working node and the in-network computing switch, and The first working node is any one of N working nodes, and the N is an integer greater than 2;

The computing switch in the network sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the congestion information, so that the N working nodes are based on The congestion information performs congestion control.
The method according to claim 1, wherein the acquiring congestion information by the in-network computing switch comprises:

The in-network computing switch obtains the first message sent by the first working node, the first message carries the congestion information, and the congestion information includes the display congestion notification ECN flag bit of the first message The first value on the above, the first value is used to indicate that the first communication link is congested;

Correspondingly, the in-network computing switch sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the congestion information, including:

During the congestion indication aging period of the first value, the in-network computing switch sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the Congestion information.
The method according to claim 2, wherein before the in-network computing switch sends N packets with the same sequence number to the N working nodes, the method further comprises:

The in-network computing switch modifies the value of the ECN flag bit of the second packet to the first value to obtain a third packet, and the second packet is the first aggregation within the validity period. For a completed message, the sequence number of the second message is the same as the sequence number of the third message;

Correspondingly, the in-network computing switch sending N packets with the same sequence number to the N working nodes includes:

The computing switch in the network sends N third messages to the N working nodes, the first value in each third message indicates that the corresponding working node performs congestion control, and the N The sequence numbers in the third message are the same.
The method according to any one of claims 2-3, wherein the method further comprises:

If the in-network computing switch receives the fourth packet sent by the first working node within the congestion indication aging period of the first value, the value on the ECN flag bit in the fourth packet is The first value;

The in-network computing switch ignores the first value in the fourth packet.
The method according to claim 3, wherein the in-network computing switch modifies the value of the ECN flag bit of the second packet to the first value, comprising:

When the ECN flag bit of the first packet includes the first ECN field, and the ECN flag bit of the second packet includes the second ECN field, the in-network computing switch will set the value in the second ECN field The value is modified to the first value in the first ECN field; or,

When the ECN flag bit of the first message includes the first forward display congestion notification FECN bit, and the ECN flag bit of the second message includes the second FECN bit, the in-network computing switch will The value in the second FECN bit is modified to the first value in the first FECN bit; or,

When the ECN flag bit of the first message includes the first backward display congestion notification BECN bit, and the ECN flag bit of the second message includes the second BECN bit, the in-network computing switch will The value in the second BECN bit is modified to the first value on the first BECN bit.
The method according to claim 1, wherein the acquiring congestion information by the in-network computing switch comprises:

When the port status between the first working node and the in-network computing switch shows congestion, the in-network computing switch modifies the value of the ECN flag bit in the N data messages to be broadcast to Obtain congestion information, where the N data messages to be broadcast are messages with the same sequence number among the N working nodes;

Correspondingly, the in-network computing switch sends N packets with the same sequence number to the N working nodes, and the N packets with the same sequence number all carry the congestion information, including:

The in-network computing switch sends the modified N data messages to be broadcast to the N working nodes, and the values on the ECN flag bits in the modified N data messages to be broadcast are used respectively Instruct the N working nodes to perform congestion control.
The method according to claim 6, wherein the in-network computing switch modifies the value of the ECN flag bit in the N data messages to be broadcast, comprising:

When the ECN flag in the data message to be broadcast includes the third ECN field, the in-network computing switch sets the value in the third ECN field; or,

When the ECN flag bit in the data message to be broadcast includes the third FECN bit, the in-network computing switch sets the value in the third FECN bit.
An in-network computing switch, which is characterized in that it comprises:

The acquiring unit is configured to acquire congestion information, where the congestion information is used to indicate that the first communication link is congested, and the first communication link is the link between the first working node and the computing switch in the network, so The first working node is any one of N working nodes, and the N is an integer greater than 2;

The sending unit is configured to send N messages with the same sequence number to the N working nodes, and the N messages with the same sequence number all carry the congestion information, so that the N working nodes are based on all the messages. The congestion information is described for congestion control.
The in-network computing switch according to claim 8, wherein the acquiring unit comprises:

The first obtaining module is configured to obtain a first message sent by the first working node, the first message carrying the congestion information, and the congestion information includes the display congestion notification ECN flag of the first message A first value on the bit, where the first value is used to indicate that the first communication link is congested;

Correspondingly, the sending unit includes:

The first sending module is configured to send N packets with the same sequence number to the N working nodes within the congestion indication time period of the first value obtained by the first obtaining module, and the N All packets with the same sequence number carry the congestion information.
The in-network computing switch according to claim 9, wherein the in-network computing switch further comprises:

The modifying unit is configured to modify the value of the ECN flag bit of the second message to the first value before the first sending module sends N messages with the same sequence number to the N working nodes, To obtain a third message, where the second message is the first message to be aggregated within the validity period, and the sequence number of the second message is the same as the sequence number of the third message;

Correspondingly, the first sending module includes:

The first sending submodule is configured to send the N third messages obtained by the modifying unit to the N working nodes, and the first value in each third message indicates the corresponding work The node performs congestion control, and the sequence numbers in the N third messages are the same.
The in-network computing switch according to any one of claims 9-10, wherein the in-network computing switch further comprises:

The acquiring unit is configured to receive a fourth message sent by the first working node within the congestion indication aging period of the first value, and the value on the ECN flag bit in the fourth message Is the first value;

The ignoring unit is configured to ignore the first value in the fourth message obtained by the obtaining unit.
The in-network computing switch according to claim 10, characterized in that:

The modification unit is configured to, when the ECN flag bit of the first packet includes a first ECN field, and the ECN flag bit of the second packet includes a second ECN field, add the second ECN field to The value of is modified to the first value in the first ECN field; or,

The modification unit is configured to: when the ECN flag bit of the first message includes the first forward display congestion notification FECN bit, and the ECN flag bit of the second message includes the second FECN bit, then the The value in the second FECN bit is modified to the first value in the first FECN bit; or,

The modification unit is configured to: when the ECN flag bit of the first message includes the first backward display congestion notification BECN bit, and the ECN flag bit of the second message includes the second BECN bit, then the The value in the second BECN bit is modified to the first value on the first BECN bit.
The in-network computing switch according to claim 8, wherein the acquiring unit comprises:

The second acquisition module is configured to modify the value of the ECN flag bit in the N data messages to be broadcast when the port status between the first working node and the in-network computing switch shows congestion, To obtain congestion information, wherein the N data messages to be broadcast are messages with the same sequence number among the N working nodes, and the second working node is any one of the N working nodes;

Correspondingly, the sending unit includes:

The second sending module is configured to send the modified N data messages to be broadcast to the N working nodes, and the values on the ECN flag bits in the modified N data messages to be broadcast are respectively used Instruct the N working nodes to perform congestion control.
The in-network computing switch according to claim 13, characterized in that:

The second acquiring module is configured to, when the ECN flag bit in the data message to be broadcast includes a third ECN field, the in-network computing switch sets the value in the third ECN field; or,

The second acquiring module is configured to, when the ECN flag bit in the data message to be broadcast includes the third FECN bit, the in-network computing switch sets the value in the third FECN bit.
A computer device, characterized by comprising: a processor, the processor is coupled with a memory, the memory is used to store a program or instruction, when the program or instruction is executed by the processor, the computer The device executes the method according to any one of claims 1 to 7.
A computer-readable storage medium having a computer program or instruction stored thereon, wherein the computer program or instruction is executed to cause a computer to execute the method according to any one of claims 1 to 7.
A chip, characterized by comprising: a processor, the processor is coupled with a memory, the memory is used to store a program or an instruction, when the program or an instruction is executed by the processor, the network computing switch Perform the method according to any one of claims 1 to 7.