Packet sending method, global scheduling ethernet network, electronic device, and storage medium

Abstract

The present application provides a packet sending method, a global scheduling Ethernet network, an electronic device, and a storage medium. The packet sending method comprises: a first global scheduling processor (GSP) node in a global scheduling Ethernet network determining a target uplink from among a plurality of uplinks on the basis of first information, the plurality of uplinks being links between the first GSP node and a plurality of global scheduling fabrics (GSFs) in the global scheduling Ethernet network; and using the target uplink to send a packet container (PKTC). When GSP nodes in the global scheduling Ethernet network send PKTCs, appropriate uplinks can be selected on the basis of the first information to send the PKTCs.

Description

Message sending method, full-scheduling Ethernet network, electronic device and storage medium

[0001] Cross-reference

[0002] The present application claims priority to the Chinese patent application No. 202410441379.3, filed on April 12, 2024, and entitled "Message sending method, full-scheduling Ethernet network, electronic device and storage medium", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0003] The present application relates to the field of network communication technology, and in particular to a message sending method, a full-scheduling Ethernet network, an electronic device and a storage medium. BACKGROUND

[0004] Full-scheduling Ethernet is a new type of Ethernet architecture, which aims to break through the performance bottleneck of the intelligent center network, create a new intelligent center network with no blocking, high bandwidth and ultra-low latency, and better serve high-performance computing such as Artificial Intelligence-Generated Content (AIGC), and meet the demand of large-scale deployment and calculation of AI. The network architecture of full-scheduling Ethernet can be divided into control layer, network layer and computing layer from top to bottom. One or more Global Scheduling Operating System (GSOS) nodes are deployed in the control layer. A plurality of Global Scheduling Processor (GSP) nodes and a plurality of Global Scheduling Fabirc (GSF) nodes are deployed in the network layer. A plurality of computing nodes are deployed in the computing layer. When transmitting data, the GSP node in the network layer can send the data to be transmitted to the GSF node, and the GSF node forwards the data to other GSP nodes.

[0005] In the case of serving AI training and inference tasks, the data traffic generated by the full-scheduling Ethernet network usually has the characteristics of periodicity, burstiness, elephant flow and low entropy. Therefore, when the data is forwarded by the GSF node, congestion problems inevitably occur at the outlet of the GSF node. In order to improve the congestion problem, in the related art, a load balancing mechanism of a packet container (PKTC) is usually used, that is, when the GSP node sends data to the GSF node, the GSP node can send a fixed-length PKTC, so as to avoid the congestion problem caused by the continuous convergence of the elephant flow at the GSF outlet. However, although the load balancing mechanism of the PKTC can avoid the continuous convergence of the traffic at the GSF outlet, there is still a cache backlog problem at the outlet of the GSF node. For example, when multiple GSP nodes send PKTCs to the GSF node, the GSP nodes may send PKTCs to the same GSF node at the same time. As a result, multiple PKTCs are accumulated at the outlet of the GSF node, which causes a long queuing delay and affects the task completion time and business efficiency. SUMMARY

[0006] The present application provides a packet sending method, a full-scheduling Ethernet network, an electronic device and a storage medium.

[0007] In a first aspect, a packet sending method is provided, which is applied to a first full-scheduling network processing node GSP in a full-scheduling Ethernet network. The method comprises: determining a target uplink from a plurality of uplinks according to first information, the plurality of uplinks being links between the first GSP and a plurality of full-scheduling switching network GSFs in the network; and sending a packet container PKTC using the target uplink.

[0008] In a second aspect, a full-scheduling Ethernet network is provided, which comprises a control layer, a network layer and a computing layer. The network layer comprises a plurality of GSFs and a plurality of GSPs. For any first GSP in the plurality of GSPs, the first GSP is configured to determine a target uplink from a plurality of uplinks according to first information, and send a PKTC using the target uplink, the plurality of uplinks being links between the first GSP and the plurality of GSFs.

[0009] In a third aspect, an electronic device is provided, which comprises: a processor; a memory for storing instructions executable by the processor; and wherein the processor is configured to execute the instructions to implement the method of the first aspect.

[0010] In a fourth aspect, a computer-readable storage medium is provided, which, when the instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the method of the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only constitute some embodiments described in the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0012] FIG. 1 is a schematic diagram of a network architecture of a full-scheduling Ethernet in the related art;

[0013] FIG. 2 is a schematic diagram of a GSP sending a PKTC in the related art;

[0014] FIG. 3 is a schematic diagram of a packet sending method according to an embodiment of the application;

[0015] FIG. 4 is a schematic diagram of indicating an uplink through a signaling packet according to an embodiment of the application;

[0016] FIG. 5 is a schematic diagram of an encapsulation manner of a signaling packet according to an embodiment of the application;

[0017] FIG. 6 is a schematic diagram of indicating an uplink through a signaling packet according to an embodiment of the application;

[0018] FIG. 7 is a schematic diagram of indicating an uplink through a signaling packet according to an embodiment of the application;

[0019] FIG. 8 is a schematic diagram of the structure of an electronic device according to an embodiment of the application;

[0020] FIG. 9 is a schematic diagram of the structure of a packet sending apparatus according to an embodiment of the application. DETAILED DESCRIPTION

[0021] An AI center network is a network that can run AI training and inference tasks, usually adopts an advanced CLOS architecture, and adopts a low-convergence-ratio or even non-blocking fabric network design. The non-blocking of the AI fabric network means that it can support full-duplex between all pairs of ports, but it does not mean that there will be no congestion in the network. For example, if there is a persistent multi-to-one, it will still cause packet squeezing at the node's egress, and once the cache threshold is exceeded, congestion will occur at the node's egress. In order to solve the congestion problem, the industry proposes a network architecture of a full-scheduling Ethernet (which can be represented as Global Scheduled Ethernet, GSE for short), which aims to break through the performance bottleneck of the AI center network and create a new type of AI center network with non-blocking, high bandwidth and ultra-low latency, in order to better serve AIGC and other high-performance computing, and meet the demand for large-scale deployment and calculation of AI.

[0022] FIG. 1 is a schematic diagram of a network structure of a full-scheduling Ethernet in the related art. As shown in FIG. 1, the network architecture of the full-scheduling Ethernet can be divided into a control layer 11, a network layer 12 and a computing layer 13 from top to bottom. A plurality of (or one) GSOS nodes are deployed in the control layer 11, and the GSOS node is a control plane component and can provide centralized network operating system capabilities for network management. A plurality of GSP nodes and a plurality of GSF nodes are deployed in the network layer 12, the GSP node is a leaf node, used to access computing traffic and perform global scheduling on the traffic, and the GSF node is a center node (spine node) above the GSP node, used to flexibly expand the network scale, and has dynamic load balancing capability and back pressure information publishing capability. A plurality of computing nodes are deployed in the computing layer 13, and the computing node can be a server-side computing card, a network card, etc., used to provide high-performance computing capability. When transmitting data, the GSP node in the network layer 11 can forward the data to other GSP nodes through the GSF node.

[0023] The network architecture shown in FIG. 1, when serving AI training and inference tasks, has a congestion problem at the egress of the GSP node and the GSF node due to the periodic, bursty and elephant flow, low entropy characteristics of the data traffic generated. In the related art, to solve the egress congestion problem of the GSP node and the GSF node, for the GSP node, an end-to-end authorization mechanism based on a dynamic global scheduling queue (DGSQ) can be used, and the source GSP backlog traffic is monitored, and after exceeding the limit, a priority-based flow control (PFC) flow control is used to notify the source to slow down. For the GSF node, a load balancing mechanism based on PKTC can be used to avoid the congestion caused by the continuous convergence of elephant flow at the GSF egress due to the traditional flow level equal cost multi path (ECMP).

[0024] However, in actual application, for the GSF node, although the load balancing mechanism based on the PKTC can avoid the continuous convergence of the GSF export traffic, there is still a cache backlog problem. For example, in some scenarios, multiple GSP nodes may choose to send to the same uplink when sending traffic to the GSF node, that is, send the PKTC to the same GSF node at the same time, which will cause multiple PKTCs to accumulate at the export of the GSF node (in the worst case, there will be N-1 PKTC backlog, N is the number of GSP nodes). Taking the two GSF nodes and four GSP nodes shown in FIG. 2 as an example, GSP1, GSP2, and GSP3 send PKTC to GSP4 through GSF, if the three GSPs all choose to send PKTC on the first link at the same time, that is, send PKTC to GSF1, then 3 PKTC caches will be caused at the export of GSF1. In related technologies, multiple GSPs may independently select uplinks when sending PKTC according to different predetermined strategies, for example, selecting in order from small to large according to the port number, or independently selecting the export according to the current uplink load (link utilization, cache occupancy), etc. Due to the periodicity and burstiness of the intelligent computing center business, the possibility of GSF cache backlog of PKTC is increased.

[0025] In the case of PKTC accumulation at the export of the GSF node, the queuing time of the PKTC will be prolonged, and the flow completion time (FCT) of the traffic will also be increased, which will increase the single iteration task of AI training and the job completion time (JCT) time, and seriously affect the business efficiency. For example, for a typical large-scale intelligent computing center network, the number of GSP nodes may reach thousands. Taking 1000 GSP nodes, 50 GSF nodes, and a PKTC size of 4KByte as an example, if the 1000 GSF nodes send PKTC to one of the GSFs at the same time, the theoretical backlog at the export of the GSF may reach: 1000*4KByte=4MByte. If the GSF port rate is 100Gbps, the queuing delay is: 4MByte*8 / 100Gbps=800us. Even if the number of GSP nodes is only 100, the queuing delay is 80us. In the case of long queuing delay, the flow completion time of the traffic will be increased, thereby increasing the job completion time and reducing the business efficiency.

[0026] The embodiment of the present application provides a message sending method, a full-scheduling Ethernet network, an electronic device and a storage medium. For each GSP in the full-scheduling Ethernet network, a target uplink for carrying a PKTC can be determined according to first information before the PKTC is sent, and then the PKTC is sent by using the target uplink. In this way, since the GSP can select a suitable uplink to send the PKTC according to the first information when the PKTC is sent, the GSPs can be prevented from sending the PKTCs to the same GSF at the same time, so that the number of the PKTCs sent to the same GSF at the same time can be reduced, the backlog of the PKTCs at the exit of the GSF node can be reduced, the task completion time can be reduced, and the service efficiency can be improved.

[0027] In order for those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the present application will be clearly and completely described below in combination with the drawings in one or more embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the protection scope of the present application.

[0028] The terms "first", "second", and the like in the present application and claims are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the present application can be implemented in an order other than that illustrated or described herein. In addition, "and / or" in the present application and claims indicates at least one of the connected objects, and the character " / " generally indicates that the front and rear associated objects are in an "or" relationship.

[0029] The technical solutions provided by the embodiments of the present application will be described in detail below in combination with the drawings.

[0030] FIG. 3 is a flow diagram of a message sending method according to an embodiment of the present application. The method shown in FIG. 3 is applied to a first GSP in a full-scheduling Ethernet network, and the network architecture of the full-scheduling Ethernet network can be as shown in FIG. 1. The first GSP can be any GSP in FIG. 1. The method shown in FIG. 3 includes the following steps.

[0031] Step S302: determining a target uplink from a plurality of uplinks according to first information, the plurality of uplinks being links between the first GSP and a plurality of GSFs in the network.

[0032] Step S304: sending the PKTC by using the target uplink.

[0033] In the full-scheduling Ethernet network, when sending the PKTC to the GSF, the first GSP in the network layer can first determine a target uplink from the multiple uplinks according to the first information, and then send the PKTC using the target uplink, that is, select a suitable uplink from the multiple uplinks according to the first information to send the PKTC. The multiple uplinks are links between the first GSP and multiple GSFs in the full-scheduling Ethernet network. For example, if the GSFs in the network are GSF1, GSF2 and GSF3 respectively, then the multiple uplinks between the first GSP and the GSFs are the uplink between the first GSP and GSF1, the uplink between the first GSP and GSF2, and the uplink between the first GSP and GSF3. The first GSP can determine a target uplink from the multiple uplinks according to the first information, which can be that the first GSP selects a target GSF from the multiple GSFs according to the first information, and correspondingly, the first GSP sends the PKTC using the target uplink, which can be that the first GSP sends the PKTC to the target GSF, and the link between the target GSF and the first GSP is the target uplink. For example, assuming that the target uplink selected by the first GSF is the uplink between the first GSP and GSF2, then when sending the PKTC, the first GSP specifically sends the PKTC to GSF2.

[0034] In this way, since the first GSP can determine a target uplink for carrying the PKTC according to the first information before sending the PKTC, and then send the PKTC using the target uplink, it can be avoided that multiple GSPs send the PKTC to the same GSF at the same time, so as to reduce the number of PKTCs sent to the same GSF at the same time, and further reduce the backlog of PKTCs at the GSF node exit, finally reduce the task completion time and improve the business efficiency. For example, taking the network including 1000 GSP nodes and 50 GSF nodes as an example, if each GSP can evenly send the PKTC to different GSFs for carrying, then the number of PKTCs accumulated on each GSF is at most 1000 / 50=20, and in the case that the GSF port rate is 100Gbps, the queuing delay is reduced to 1.6us.

[0035] The above first information can be auxiliary information for assisting the first GSP to determine the target uplink from the multiple uplinks. The first information can be determined by the first GSP itself, or can be sent to the first GSP by other GSPs, or GSFs, or GSOS in the network, which is not limited here.

[0036] In some embodiments, the first information can include at least one of the following: a first signaling message, the first signaling message being used to indicate a first link identifier of an uplink used by the second GSP to send the PKTC; a preset link selection algorithm and a pseudo-random number, the link selection algorithm being used to output a link identifier according to input parameters; a third signaling message, the third signaling message being used to indicate a target link identifier, the target link identifier being a link identifier of a target uplink.

[0037] The second GSP can be any one of the other GSPs in the network except the first GSP. In an embodiment, the second GSP can be a GSP adjacent to the first GSP. The first link identifier can be a link number or other information that can be used to distinguish different uplinks, which is not specifically limited here. The link selection algorithm is an algorithm strategy for selecting an uplink, which can output a link identifier according to input parameters. In an embodiment, the link selection algorithm can be an existing algorithm strategy for selecting an uplink, such as selecting in order from small to large according to the port number, or selecting according to the current uplink load (link utilization, cache occupancy, congestion level information), etc. The pseudo-random number can be flexibly set, such as the local clock at the current stage, etc., which is not specifically limited here.

[0038] When the first GSP determines the target uplink from the plurality of uplinks according to the first information, the first GSP can determine the target uplink according to one or more of the above three first information, which is not specifically limited here. The following will be described in detail with respect to each first information.

[0039] In the case where the first information includes a first signaling message, the first signaling message being used to indicate a first link identifier of an uplink used by the second GSP to send the PKTC, the first GSP determining the target uplink from the plurality of uplinks according to the first information can include: determining a second link identifier according to the first link identifier indicated by the first signaling message, the second link identifier being different from the first link identifier; determining the uplink corresponding to the second link identifier as the target uplink.

[0040] Specifically, since the first signaling message indicates the link identifier of the uplink used by the second GSP to send the PKTC, the first GSP can know which GSF the second GSP sends the PKTC to according to the indication of the first signaling message. In this way, the first GSF can select another uplink when selecting the target uplink, i.e., selecting to send the PKTC to another GSF, thereby avoiding the problem of PKTC backlog caused by the second GSP sending the PKTC to the same GSF at the same time. When selecting the target uplink, the first GSP can first determine a second link identifier different from the first link identifier indicated in the first signaling message, and then determine the uplink corresponding to the second link identifier as the target uplink. Since the second link identifier is different from the first link identifier, it can be ensured that the target uplink selected by the first GSP is different from the uplink selected by the second GSP.

[0041] Taking GSFs in the network as GSF1, GSF2 and GSF3 for example, assuming that the link identifier between the GSP and GSF1 is 1, the link identifier between the GSP and GSF2 is 2, and the link identifier between the GSP and GSF3 is 3, if the first signaling message indicates that the link identifier of the uplink used by the second GSP to send the PKTC is 2, i.e., the second GSP selects to send the PKTC to GSF2, the first GSP can determine the uplink with link identifier 1 or 3 as the target uplink, i.e., the first GSP can select to send the PKTC to GSF1 or GSF3.

[0042] It should be noted that when determining the second link identifier, the second link identifier can be different from the first link identifier indicated in the first signaling message, and can also be different from the link identifier of the uplink used by the first GSP to send the PKTC last time, so as to avoid the problem of congestion of the first GSP egress caused by the first GSP using the same uplink to send the PKTC twice. For example, still taking the first signaling message indicating that the link identifier of the uplink used by the second GSP to send the PKTC is 2 as an example, when selecting the target uplink, if the link identifier of the uplink used by the first GSP to send the PKTC last time is 1, the first GSP can determine the uplink with link identifier 3 as the target uplink, which is different from the uplink selected by the second GSP and also different from the uplink used by the first GSP to send the PKTC last time.

[0043] In some embodiments, the first signaling message can be sent to the first GSP by a second GSP in the network or a GSF in the network. In this way, before the first GSP determines the target uplink from the plurality of uplinks according to the first signaling message, the first GSP can include any of the following: receiving the first signaling message sent by the second GSP; receiving the first signaling message sent by the first GSF, the first signaling message being sent by the second GSP to the first GSF.

[0044] Specifically, after the second GSP selects the uplink used to send the PKTC, the second GSP can generate a first signaling message indicating the link identifier of the uplink, and send the first signaling message directly to the first GSP, or the second GSP can also send the first signaling message to a first GSF in the network (which can be any GSF in the network), and then the first GSF forwards the first signaling message to the first GSP. In this way, the first GSP can receive the first signaling message from the second GSP or the first GSF. The second GSP can be a GSP adjacent to the first GSP.

[0045] It should be noted that in order to facilitate the first GSP to receive the first signaling message in time, and to avoid the delay of the first signaling message due to the accumulation of service messages, the first signaling message can have the highest transmission priority, so that the transmission delay of the first signaling message between the second GSP and the first GSP, or the transmission delay between the second GSP to the first GSF and the first GSF to the first GSP, can be controlled at the us level. For the case where the first signaling message is forwarded to the first GSP by the first GSF, since the first signaling message has the highest transmission priority, the queuing delay of the first signaling message at the first GSF exit can be ignored, and the physical link delay can be considered as a relatively fixed value, so the transmission delay of the first signaling message can be theoretically calculated.

[0046] After the first GSP determines the target uplink from the plurality of uplinks according to the first signaling message, in an embodiment, in some embodiments, the first GSP can also perform the following operation: sending a second signaling message to a third GSP.

[0047] After the first GSP determines the target uplink, the first GSP can generate a second signaling message, and then send the second signaling message to a third GSP, the second signaling message being used to indicate a second link identifier, the second link identifier being the link identifier of the target uplink selected by the first GSP according to the first signaling message. After receiving the second signaling message, the third GSP can determine the target uplink used to send the PKTC according to the indication of the second signaling message. The specific implementation can refer to the specific implementation of the first GSP determining the target uplink according to the indication of the first signaling message, which will not be described in detail here. The third GSP can be a GSP adjacent to the first GSP.

[0048] Since the first GSP can inform the third GSP of the selected target uplink through the second signaling message, the third GSP can select a suitable uplink to send the PKTC, and the problem of PKTC backlog caused by the first GSP and the third GSP simultaneously sending the PKTC to the same GSF can be avoided.

[0049] In an embodiment, when the first GSP sends the second signaling message to the third GSP, the second signaling message can be sent directly to the third GSP, or the second signaling message can be forwarded to the third GSP through a certain GSF in the network, which is not specifically limited here. In order to facilitate the third GSP to receive the third signaling message in time, the third signaling message can have the highest transmission priority.

[0050] In some embodiments, when the third GSP is a neighboring GSP of the first GSP, before the first GSP sends the second signaling message to the third GSP, the method can further include: determining the third GSP from the plurality of GSPs according to the second information, the second information including at least one of IP addresses, MAC addresses, and identification information of the plurality of GSPs.

[0051] That is, when the first GSP sends the second signaling message to the neighboring third GSP, the first GSP needs to determine the third GSP from the plurality of GSPs in the network first, and then send the second signaling message to the third GSP. When determining the third GSP from the plurality of GSPs, at least one of the IP addresses, the MAC addresses, and the identification information of the plurality of GSPs can be used for determination. For example, the IP loopback address of the first GSP is 10.0.0.1, and the IP addresses of the other two GSP nodes are 10.0.0.2 and 10.0.0.3, respectively. It can be considered that the GSP node with the number 10.0.0.2 is adjacent to the first GSP node, and the GSP node corresponding to 10.0.0.2 is considered as the third GSP node adjacent to the first GSP.

[0052] In a more specific embodiment based on the above-described method, for each GSP in the network, after the first signaling message determines the target uplink, the selected link identifier of the target uplink can be notified to the adjacent GSP node through the second signaling message, so that each GSP in the network can select a suitable uplink to send the PKTC, thereby avoiding the problem of PKTC backlog caused by multiple GSPs in the network simultaneously sending PKTC to the same GSF. In an embodiment, in order to balance the sending of PKTC from multiple GSPs to multiple GSFs (i.e., the number of PKTC received by each GSF is basically the same), each GSP can also select according to a certain rule when selecting the uplink according to the first signaling message (or the second signaling message). For example, in the case of link identifier being link number, after receiving the signaling message, each GSP node can perform X (X is greater than or equal to 1) processing on the link identifier indicated by the signaling message, and take the uplink corresponding to the link number obtained after the X processing as the target uplink. When the link number exceeds the number of GSF nodes (the number of uplinks), take the link number modulo the number of GSF nodes. Taking a network including three GSFs and uplink numbers 1, 2, and 3 as an example, assuming that GSP1 notifies the adjacent GSP2 node through the signaling message that the link number of the selected uplink is 1, then GSP2 can select the uplink corresponding to the link number 2 (performing 1 processing on the link number in the signaling message) after receiving the signaling message, and then notify the adjacent GSP3 node through the signaling message that the link number of the selected uplink is 2. GSP3 can select the uplink corresponding to the link number 3 after receiving the signaling message, and then notify the adjacent GSP4 node through the signaling message that the link number of the selected uplink is 3. GSP4 can select the uplink corresponding to the link number 1 after receiving the signaling message, and then notify the adjacent GSP5 node through the signaling message that the link number of the selected uplink is 1. In this way, the last GSP node selects a certain uplink according to the signaling message.

[0053] In some embodiments, in the case where the first information includes a preset link selection algorithm and a pseudo-random number, the first GSP determines the target uplink from the plurality of uplinks according to the first information, which can include: taking the pseudo-random number as an input parameter of the link selection algorithm, determining a third link identifier output by the link selection algorithm; and determining the uplink corresponding to the third link identifier as the target uplink.

[0054] The pseudo-random number is taken as an input parameter of the link selection algorithm, which can be added as an algorithm input in the link selection algorithm, and the uplink is independently selected in combination with a Hash routing mechanism. The Hash range can be 0-N-1, and N is the number of GSF nodes in the network. It should be noted that the pseudo-random number used in the embodiments of the present application is not specifically limited, and the Hash algorithm used is also not specifically limited. The first GSP can select the uplink based on the link selection algorithm and the pseudo-random number when performing link switching after the current link PKTC has been sent.

[0055] Taking a network including five GSP nodes and three GSF nodes as an example, the pseudo-random number used by each GSP and the uplink selected according to the pseudo-random number and the link selection algorithm can be as shown in Table 1.

[0056] Table 1

[0057] Based on Table 1, when the uplink is selected according to the pseudo-random number and the link selection algorithm, the number of the uplink selected by GSP1 is 1, that is, GSP1 selects to send PKTC to GSF1, the number of the uplink selected by GSP2 is 3, that is, GSP2 selects to send PKTC to GSF3, the number of the uplink selected by GSP3 is 2, that is, GSP3 selects to send PKTC to GSF2, the number of the uplink selected by GSP4 is 2, that is, GSP4 selects to send PKTC to GSF2, and the number of the uplink selected by GSP5 is 1, that is, GSP5 selects to send PKTC to GSF1. In this way, the five GSPs can send PKTC to the three GSFs evenly, avoiding the problem of PKTC backlog caused by sending PKTC to the same GSF.

[0058] In an embodiment, in some implementations, after the first GSP takes the pseudo-random number as an input parameter of the link selection algorithm and determines the third link identifier output by the link selection algorithm, the first GSP can perform the following operations: determining whether the third link identifier is the same as a fourth link identifier, the fourth link identifier being a link identifier of an uplink used by the first GSP when the first GSP last sent PKTC; in the case where the third link identifier is the same as the fourth link identifier, determining a fifth link identifier according to the third link identifier, the fifth link identifier being different from the third link identifier; and determining the uplink corresponding to the fifth link identifier as the target uplink.

[0059] Specifically, after determining the third link identifier output by the link selection algorithm based on the pseudo-random number, the first GSP can determine whether the third link identifier is the same as the link identifier of the uplink used by the first GSP when the first GSP last sent the PKTC (i.e., the fourth link identifier). If not, the first GSP can determine the uplink corresponding to the third link identifier as the target uplink. If so, the first GSP can select another uplink to avoid the problem of congestion at the first GSP egress caused by the first GSP using the same uplink to send the PKTC twice in succession. When selecting another uplink, the first GSP can determine a fifth link identifier different from the third link identifier (i.e., the fourth link identifier) and determine the uplink corresponding to the fifth link identifier as the target uplink. In determining the fifth link identifier, the method can have various forms, such as incrementing the third link identifier, etc., as long as the fifth link identifier is different from the third link identifier. Here, how to determine the fifth link identifier is not limited.

[0060] For example, the first GSP determines the link identifier as 2 based on the pseudo-random number and the link selection algorithm. If the link identifier of the uplink used by the first GSP when the first GSP last sent the PKTC is 1, the first GSP can determine the uplink corresponding to the link identifier 2 as the target uplink. If the link identifier of the uplink used by the first GSP when the first GSP last sent the PKTC is 2, the first GSP can determine the uplink corresponding to the link identifier 3 (incrementing the link identifier 2) as the target uplink to avoid congestion at the first GSP egress.

[0061] In some embodiments, in the case where the first information includes a third signaling message, the third signaling message being used to indicate a target link identifier, the first GSP determining the target uplink from the plurality of uplinks according to the first information can include: determining the uplink corresponding to the target link identifier as the target uplink.

[0062] The target link identifier is the link identifier of the target uplink. Since the third signaling message directly indicates the link identifier of the target uplink, the first GSP can directly determine the uplink corresponding to the link identifier indicated by the third signaling message as the target uplink.

[0063] It should be noted that the target link identifier indicated by the third signaling message can be the same as or different from the link identifier of the uplink used by the first GSP when the first GSP last sent the PKTC. If the same, it indicates that the first GSP needs to use the same uplink to send the PKTC twice in succession, and in this case, the first GSP egress can have a congestion problem. To avoid the congestion problem, in an embodiment, in some embodiments, when the first GSP determines the target uplink according to the third signaling message, the first GSP can also perform the following operations:

[0064] determining whether the target link identifier is the same as a fourth link identifier, the fourth link identifier being a link identifier of an uplink used by the first GSP when the first GSP last sent the PKTC; and determining the target uplink according to a preset link selection rule in a case where the target link identifier is the same as the fourth link identifier.

[0065] Specifically, when determining the target uplink according to the third signaling packet, the first GSP can first determine whether the target link identifier indicated by the third signaling packet is the same as a link identifier of an uplink used by the first GSP when the first GSP last sent the PKTC. If not, the first GSP can directly determine the uplink corresponding to the target link identifier as the target uplink. If the target link identifier is the same as the link identifier of the uplink, the first GSP can select another uplink as the target uplink in order to avoid congestion at the exit of the first GSP. When selecting the other uplink, the first GSP can select the other uplink according to a preset link selection rule. The preset link selection rule can be an existing link selection strategy, such as selecting the uplinks in order from small to large according to the port number, or selecting the uplinks according to the current uplink load (link utilization rate, cache occupancy rate, congestion level information), and the like. Alternatively, the first GSP can randomly select an uplink as the target uplink, as long as the selected uplink is different from the uplink indicated by the third signaling packet. The manner of selecting the other uplink is not limited herein.

[0066] In some embodiments, the third signaling packet can be sent to the first GSP by another GSP in the network, or by a certain GSF in the network, or by a certain GSOS in the network. In this way, before determining the target uplink from the multiple uplinks according to the third signaling packet, the first GSP can include any one of the following: receiving the third signaling packet sent by a fourth GSP; receiving the third signaling packet sent by a second GSF, the third signaling packet being sent by the fourth GSP to the second GSF; and receiving the third signaling packet sent by a GSOS.

[0067] The fourth GSP can be any one of the GSPs in the network except the first GSP. In an embodiment, the fourth GSP can be a designated GSP in the network, which has the authority to uniformly control and configure each GSP node in the network, and can uniformly manage and allocate the uplinks used by each GSP node in the network when sending the PKTC and configure the uplinks to the GSP nodes in a global perspective. When configuring the third signaling packet to the first GSP, the fourth GSP can directly send the third signaling packet to the first GSP, or the fourth GSP can forward the third signaling packet to the first GSP through a GSF in the network, that is, the fourth GSP first sends the third signaling packet to a second GSF (which can be any GSF in the network) in the network, and then the second GSF forwards the third signaling packet to the first GSP. In this way, the first GSP can receive the third signaling packet from the fourth GSP or the second GSF.

[0068] In an embodiment, in addition to indicating the link identifier, the third signaling message of the fourth GSP configuration can also carry information such as the type of the signaling message and the ID of the destination GSP. For example, the fourth GSP can query the IDs of other GSP nodes according to a local table, and then can respectively specify different uplink links for different GSP nodes to send PKTC in a certain rule (for example, incrementally).

[0069] The GSOS can be any GSOS in the network. The GSOS node is used for the overall management and configuration of the entire network, and the GSOS needs to configure the DGSQ authorization value of each GSP node, so it naturally has the management capability of the GSP nodes in the entire network, and in order to make the authorization take effect quickly, the link between the GSOS node and the GSP node generally adopts out-of-band networking or uses a high-priority message to carry the authorization message in the network, and its response time reaches near real-time. Based on the management and configuration authority of the GSOS to the GSP nodes in the entire network, the GSOS can uniformly manage and allocate the uplink links used by each GSP node to send PKTC in the network and configure them to the GSP nodes. When the GSOS configures the third signaling message to the first GSP, the GSOS can send the third signaling message to the first GSP through out-of-band networking, so that the first GSP can receive the third signaling message from the GSOS. Alternatively, the GSOS can also forward the third signaling message to the first GSP through the GSF in the network, and the way in which the GSOS sends the third signaling message is not limited here.

[0070] In the case where the fourth GSP or the second GSF or the GSOS sends the first signaling message to the first GSP, in order to facilitate the first GSP to receive the third signaling message in time, and avoid the delay of the first signaling message due to the accumulation of service messages, the third signaling message can have the highest transmission priority.

[0071] It should be noted that the fourth GSP or the GSOS can configure different link identifiers in each round of configuration when configuring the third signaling message to the first GSP, that is, the fourth GSP or the GSOS can flexibly determine the uplink link used by the first GSP to send PKTC each time based on the current network statistical information.

[0072] The above detailed how the first GSP determines the target uplink from the multiple uplink links according to the first information, and in the case where the first information includes a signaling message (the first signaling message or the third signaling message), the first information can be periodically sent to the first GSP, for example, the first signaling message or the third signaling message can be sent to the first GSP at a certain frequency.

[0073] In general, the role of the first information can be equivalent to a "synchronization" signal between GSP nodes, and each GSP node can perform balanced routing based on the "synchronization" signal when sending a PKTC to a GSF, so as to avoid the PKTC backlog problem caused by sending PKTCs to the same GSF at the same time. Among them, the "synchronization" signal can be periodically synchronized, that is, not every time the PKTC is sent, the routing is performed based on the "synchronization" signal, and the routing can be performed based on the "synchronization" signal every other time period. The specific decision can be made according to the actual network scene, for example, synchronize once every 1s, which is not limited here. Among the adjacent two synchronizations, each GSP node can select the uplink according to the existing uplink selection mechanism (for example, according to the order from small to large of the port number, or according to the current uplink load (link utilization, buffer occupancy, congestion level information) and other conditions), considering that the intelligent algorithm business has the elephant flow feature, the PKTC uplink switching conforms to certain regularity, and the probability of sending PKTCs of multiple GSPs to the same GSF can still be reduced, thereby reducing the total task completion time.

[0074] The embodiments of the present application improve the load balancing routing strategy of GSP based on PKTC, so that each GSP can fully utilize the number advantage of GSF nodes, reduce the number of PKTCs sent to the same GSF at the same time, thereby theoretically reducing the PKTC backlog amount of GSF nodes, ultimately reducing the AI training task completion time, and improving the business efficiency.

[0075] Based on the technical solutions provided by the embodiments of the present application, the embodiments of the present application can improve the PKTC backlog problem of the GSF export in at least the following three ways: 1. By sending signaling messages between GSP nodes to announce the selected uplink number, providing a basis for adjacent GSP nodes to correctly select the uplink (corresponding to the case that the first information includes the first signaling message); 2. By adding a Hash mechanism based on a random number in the export link selection algorithm of the GSP node, providing an indication for the link to which the subsequent PKTC needs to be switched (corresponding to the case that the first information includes a preset link selection algorithm and a pseudo-random number); 3. By implementing a global network information collection and distribution strategy in the control layer GSOS node or the designated GSP node of the network layer, and directly indicating the uplink available to each GSP node by configuring and issuing messages to each GSP node (corresponding to the case that the first information includes the third signaling message). For ease of understanding, the following will be described in three more specific embodiments.

[0076] Embodiment one:

[0077] This embodiment illustrates the method of GSP routing decision based on network layer signaling message. Please refer to FIG. 4. In the network shown in FIG. 4, there are 100 GSP nodes (numbered 1-100), 20 GSF nodes (numbered 1-20), the uplink between GSF1 and GSP n (n is 1-100) is numbered 1, the uplink between GSF2 and GSP n is numbered 2, and so on, and the uplink between GSF20 and GSP n is numbered 20. Assuming that the length of all fiber links in the network is 200 meters, the link delay is 1us, and assuming that the non-blocking processing delay of GSF node is 3us, the queuing delay is ignored, the Fabric transmission delay of GSP signaling message to the adjacent GSP node is 5us (ignoring the signaling message analysis delay). Taking the PKTC size of 4K bytes and the port rate of 40Gbps as an example, the minimum serialization delay of each PKTC is: 4K*8*10^6 / 40Gbps=0.8us, that is, under the AI elephant flow load, the path is switched at a frequency of about 0.8us. During the transmission of GSP signaling message, the current GSP path is switched about 5-6 times.

[0078] In FIG. 4, after GSP1 selects a certain uplink, it can notify the link number of the selected uplink to GSP2 through a signaling message. After GSP2 receives the signaling message, it selects a certain uplink different from the link selected by GSP1, and notifies the link number of the selected uplink to GSP3 through a signaling message. In this way, until the last GSP100 selects an uplink.

[0079] Assuming that each GSP regularly selects an uplink according to the formula (1+(n-1)*6) / 20, if GSP1 selects the uplink numbered 1, then the adjacent node GSP2 of GSP1 can select the uplink numbered 7, the adjacent node GSP3 of GSP2 can select the uplink numbered 13, and so on, and GSP100 can select the uplink numbered 15.

[0080] The encapsulation method of the signaling message is not specifically limited in this embodiment. For example, a new TP-ID field value indicating carrying a new 4-byte header can be defined behind the MAC layer, or the information type carried can also be indicated through the protocol-id field behind the IPv4 / v6 message header, as shown in FIG. 5.

[0081] It should be noted that the GSP link number selection method described in this embodiment is only for example, and is not the only implementation method. The message encapsulation method is only for example, and in actual deployment, there can be many other ways to choose, which are not specifically limited in this embodiment.

[0082] Embodiment Two

[0083] This embodiment illustrates the method of balancing flexible selection of uplink based on pseudo-random number. For the convenience of illustration, this embodiment can assume that the number of GSP nodes in the network is 5 (numbered 1-5) and the number of GSF nodes is 3 (the corresponding uplink link numbers are 1-3). When the uplink is selected for the first time based on the pseudo-random number, it is assumed that the selection results of the 5 GSPs are shown in Table 2.

[0084] Table 2

[0085] When the uplink needs to be switched, each GSP node can calculate and select a new uplink based on the pseudo-random number to send the next PKTC. It is assumed that the re-selection results of the 5 GSPs are shown in Table 3.

[0086] Table 3

[0087] In Table 3, for GSP3, since the last selected uplink number is also 1, the selected link number can be incremented by 1, that is, the uplink link number 2 is selected, to avoid the problem of GSP3 egress congestion caused by GSP3 using the same uplink to send PKTC for two consecutive times.

[0088] Embodiment Three

[0089] This embodiment illustrates the method of GSP node uplink configuration based on centralized manner. It is assumed that the network includes 100 GSP nodes (numbered 1-100) and 20 GSF nodes (numbered 1-20), the uplink number between GSF1 and GSP n (n is 1-100) is 1, the uplink number between GSF2 and GSP n is 2, and so on, and the uplink number between GSF20 and GSP n is 20. When configuring the uplink for the 100 GSP nodes, the following configuration can be made: GSP1-GSP5 nodes select the uplink link number 1, GSP6-GSP10 nodes select the uplink link number 2, GSP11-GSP15 nodes select the uplink link number 3, and so on, and GSP96-GSP100 nodes select the uplink link number 20.

[0090] If the link configuration is performed by the GSOS node, the GSOS node can indicate the selectable uplink to each GSP node through FIG. 6. In FIG. 6, after the GSOS node issues the signaling message to each GSP node, each GSP node can select the corresponding uplink to send the PKTC according to the indication of the signaling message. Specifically, GSP1-GSP5 select the uplink 1 to send the PKTC, GSP6-GSP10 select the uplink 2 to send the PKTC, …, GSP96-GSP100 select the uplink 20 to send the PKTC.

[0091] The protocol and encapsulation manner of the configuration message are not limited in the embodiment, for example, an encapsulation manner based on Netconf / yang can be as follows:

[0092] If the configuration of the uplink of each GSP node is performed by the designated GSP node, the designated GSP node sends a signaling message to other nodes to perform unified management configuration. As shown in FIG. 7, the designated GSP node is GSP1 node, GSP1 node selects the uplink 1, and indicates other GSP nodes to select which uplink through the signaling message. After GSP1 node issues the signaling message to other GSP nodes, other GSP nodes can select the corresponding uplink to send the PKTC according to the indication of the signaling message. Specifically, GSP2-GSP5 select the uplink 1 to send the PKTC, GSP6-GSP10 select the uplink 2 to send the PKTC, …, GSP96-GSP100 select the uplink 20 to send the PKTC.

[0093] It should be noted that the above-mentioned embodiments one to three are only preferred embodiments, and other changes and variations can also be made in other possible implementation manners. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

[0094] Based on the technical scheme provided in the embodiment of the present application, for each GSP in the full-scheduling Ethernet network, before sending the PKTC, the target uplink used to carry the PKTC can be determined according to the first information, and then the PKTC is sent using the target uplink. In this way, since the GSP can select the appropriate uplink to send the PKTC according to the first information when sending the PKTC, the PKTCs sent to the same GSF at the same time can be avoided, thereby the number of PKTCs sent to the same GSF at the same time can be reduced, the PKTC buffer backlog at the exit of the GSF node can be reduced, the task completion time can be reduced, and the service efficiency can be improved.

[0095] The above describes specific embodiments of the application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order and still accomplish the desired result. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous or necessary.

[0096] Figure 8 is a structural schematic diagram of an electronic device according to an embodiment of the present application. Referring to Figure 8, at the hardware level, the electronic device includes a processor, and in an embodiment, an internal bus, a network interface, and a memory. The memory can include a memory, such as a random-access memory (RAM), and can also include a non-volatile memory, such as at least one disk memory. Of course, the electronic device can also include other hardware required for business.

[0097] The processor, network interface, and memory can be connected to each other through the internal bus, which can be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one bidirectional arrow is shown in Figure 8, but it does not mean that there is only one bus or only one type of bus.

[0098] The memory is used to store programs. Specifically, the program can include program code, which includes computer operation instructions. The memory can include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0099] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs, and forms a packet sending device at the logical level. The processor executes the program stored in the memory, and is specifically configured to perform the following operations: determining a target uplink from a plurality of uplinks according to first information, the plurality of uplinks being links between a first GSP in a full-scheduling Ethernet and a plurality of GSFs in a full-scheduling Ethernet network, the first GSP being any GSP in the full-scheduling Ethernet network; and sending a packet container PKTC using the target uplink.

[0100] The method performed by the packet sending device disclosed in the embodiment shown in Fig. 8 of the present application can be applied to a processor or implemented by the processor. The processor can be an integrated circuit chip with processing capability. In the implementation process, each step of the above method can be completed by the integrated logic circuit or the instruction in the form of software in the processor. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; or a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. Each method, step and logic block diagram disclosed in the present application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the present application can be directly embodied as a hardware code processor to execute, or a combination of hardware and software modules in the code processor to execute. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium in the memory is read by the processor, and the hardware thereof is combined to complete the steps of the above method.

[0101] The electronic device can also perform the method of Fig. 3 and realize the functions of the packet sending device in the embodiment shown in Fig. 3. The present application will not be repeated here.

[0102] Of course, in addition to the software implementation, the electronic device of the present application does not exclude other implementation manners, such as logic devices or a combination of software and hardware, etc. That is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or a logic device.

[0103] The application further provides a computer readable storage medium storing one or more programs including instructions which, when executed by a portable electronic device including a plurality of application programs, enable the portable electronic device to perform the method of the embodiment shown in FIG. 3, and specifically to perform the following operation: determining a target uplink from a plurality of uplinks according to first information, the plurality of uplinks being links between a first GSP in a full-schedule Ethernet network and a plurality of GSFs in the full-schedule Ethernet network, the first GSP being any GSP in the full-schedule Ethernet network; and sending a packet container PKTC using the target uplink.

[0104] FIG. 9 is a structural schematic diagram of a packet sending device 90 according to an embodiment of the application. The device shown in FIG. 9 can be applied to a first GSP in a full-schedule Ethernet network. Referring to FIG. 9, in a software implementation, the packet sending device 90 can include a determining module 91 and a sending module 92, wherein: the determining module 91 is configured to determine a target uplink from a plurality of uplinks according to first information, the plurality of uplinks being links between the first GSP and a plurality of full-schedule switch networks GSFs in the network; and the sending module 92 is configured to send a packet container PKTC using the target uplink.

[0105] In some embodiments, the first information includes a first signaling packet used to indicate a first link identifier of an uplink used by a second GSP to send the PKTC; and the determining module 91 is configured to determine a target uplink from a plurality of uplinks according to first information, including: determining a second link identifier according to the first link identifier indicated by the first signaling packet, the second link identifier being different from the first link identifier; and determining an uplink corresponding to the second link identifier as the target uplink.

[0106] In some embodiments, the device further includes a first receiving module configured to perform any of the following operations: receiving the first signaling packet sent by the second GSP, the second GSP being adjacent to the first GSP; and receiving the first signaling packet sent by a first GSF, the first signaling packet being sent by the second GSP to the first GSF; wherein the first signaling packet has the highest transmission priority.

[0107] In some embodiments, the sending module 92 is further configured to send a second signaling packet to a third GSP, the third GSP being adjacent to the first GSP; wherein the second signaling packet is used to indicate the second link identifier, and the second signaling packet is used by the third GSP to determine an uplink used to send the PKTC.

[0108] In some embodiments, the sending module 92 further comprises, before sending the second signaling packet to the third GSP, determining the third GSP from the plurality of GSPs according to second information, the second information comprising at least one of IP address, MAC address and identification information of the plurality of GSPs.

[0109] In some embodiments, the first information comprises a preset link selection algorithm and a pseudo-random number, the link selection algorithm being used to output a link identification according to an input parameter; and the determining module 91 determines the target uplink from the plurality of uplinks according to the first information, comprising: taking the pseudo-random number as the input parameter of the link selection algorithm, and determining a third link identification output by the link selection algorithm; and determining the uplink corresponding to the third link identification as the target uplink.

[0110] In some embodiments, the determining module 91 further comprises: determining whether the third link identification is the same as a fourth link identification, the fourth link identification being the link identification of the uplink used by the first GSP when last sending a PKTC; and in the case that the third link identification is the same as the fourth link identification, determining a fifth link identification according to the third link identification, the fifth link identification being different from the third link identification; and determining the uplink corresponding to the fifth link identification as the target uplink.

[0111] In some embodiments, the first information comprises a third signaling packet, the third signaling packet being used to indicate a target link identification; and the determining module 91 determines the target uplink from the plurality of uplinks according to the first information, comprising: determining the uplink corresponding to the target link identification as the target uplink.

[0112] In some embodiments, the determining module 91 further comprises: determining whether the target link identification is the same as a fourth link identification, the fourth link identification being the link identification of the uplink used by the first GSP when last sending a PKTC; and in the case that the target link identification is the same as the fourth link identification, determining the target uplink according to a preset link selection rule.

[0113] In some embodiments, the apparatus further comprises a second receiving module, the second receiving module being used to receive any one of: the third signaling packet sent by a fourth GSP, the fourth GSP being a designated GSP in the plurality of GSPs; the third signaling packet sent by a second GSF, the third signaling packet being sent by the fourth GSP to the second GSF; and the third signaling packet sent by a global scheduling operating system (GSOS); wherein the third signaling packet has the highest transmission priority.

[0114] In some embodiments, the first information is periodically sent to the first GSP in the case that the first information comprises a signaling packet.

[0115] The packet sending apparatus 90 provided in the embodiments of the present application can also execute the method shown in FIG. 3, and realize the functions of the packet sending apparatus 90 in the embodiment shown in FIG. 3. Details are not described herein again.

[0116] The embodiments of the present application also provide a full-scheduling Ethernet network, which comprises a control layer, a network layer and a computing layer, the network layer comprises a plurality of GSFs and a plurality of GSPs. For any first GSP in the plurality of GSPs, the first GSP is configured to determine a target uplink from a plurality of uplinks according to first information, and send a PKTC using the target uplink, the plurality of uplinks being links between the first GSP and the plurality of GSFs.

[0117] The control layer in the embodiments can comprise one or more GSOSs, and the computing layer can comprise a plurality of computing nodes. The structures of the network layer and the computing layer can refer to the embodiment shown in FIG. 1. The specific implementation of each step performed by the first GSP can refer to the specific implementation of the corresponding step in the embodiment shown in FIG. 3, and the same technical effects can be achieved. Details are not described herein again.

[0118] In summary, the above only describes the preferred embodiments of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0119] The system, apparatus, module or unit described in the above embodiments can be specifically implemented by a computer chip or entity, or by a product with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0120] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0121] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or apparatus that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0122] The various embodiments in the present application are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

Claims

1. A message sending method, applied to a first fully schedulable network processing node GSP in a fully schedulable Ethernet network, the method comprising: Determining a target uplink from a plurality of uplinks according to the first information, where the plurality of uplinks are links between the first GSP and a plurality of fully scheduled switching networks GSFs in the network; The message container PKTC is sent using the target uplink.

2. The method according to claim 1, wherein the first information comprises a first signaling message, wherein the first signaling message is used to indicate a first link identifier of an uplink used by the second GSP to send the PKTC; The determining a target uplink from a plurality of uplinks according to the first information includes: Determine a second link identifier according to the first link identifier indicated by the first signaling message, where the second link identifier is different from the first link identifier; An uplink corresponding to the second link identifier is determined as the target uplink.

3. The method according to claim 2, further comprising any one of the following: receiving the first signaling message sent by the second GSP, where the second GSP is adjacent to the first GSP; receiving the first signaling message sent by the first GSF, where the first signaling message is sent by the second GSP to the first GSF; in, The first signaling message has the highest transmission priority.

4. The method according to claim 2 or 3, after determining the target uplink, the method further comprises: Sending a second signaling message to a third GSP, where the third GSP is adjacent to the first GSP; The second signaling message is used to indicate the second link identifier, and the second signaling message is used by the third GSP to determine the uplink used when sending PKTC.

5. The method according to claim 4, before sending the second signaling message to the third GSP, the method further comprises: The third GSP is determined from a plurality of GSPs according to second information, where the second information includes at least one of IP addresses, MAC addresses, and identification information of the plurality of GSPs.

6. The method according to claim 1, wherein the first information comprises a preset link selection algorithm and a pseudo-random number, wherein the link selection algorithm is configured to output a link identifier according to input parameters; The determining a target uplink from a plurality of uplinks according to the first information includes: Using the pseudo-random number as an input parameter of the link selection algorithm, and determining a third link identifier output by the link selection algorithm; An uplink corresponding to the third link identifier is determined as the target uplink.

7. The method of claim 6, further comprising: Determine whether the third link identifier is the same as a fourth link identifier, where the fourth link identifier is the link identifier of the uplink used when the first GSP last sent the PKTC; In a case where the third link identifier is the same as the fourth link identifier, determining a fifth link identifier according to the third link identifier, where the fifth link identifier is different from the third link identifier; An uplink corresponding to the fifth link identifier is determined as the target uplink.

8. The method according to claim 1, wherein the first information includes a third signaling message, and the third signaling message is used to indicate a target link identifier; and determining the target uplink from the plurality of uplinks based on the first information comprises: An uplink corresponding to the target link identifier is determined as the target uplink.

9. The method of claim 8, further comprising: Determine whether the target link identifier is the same as a fourth link identifier, where the fourth link identifier is the link identifier of the uplink used when the first GSP last sent the PKTC; In a case where the target link identifier is the same as the fourth link identifier, the target uplink is determined according to a preset link selection rule.

10. The method according to claim 8 or 9, further comprising any one of the following: receiving the third signaling message sent by a fourth GSP, where the fourth GSP is a designated GSP among the multiple GSPs; receiving the third signaling message sent by the second GSF, where the third signaling message is sent by the fourth GSP to the second GSF; Receiving the third signaling message sent by the full scheduling operating system GSOS; in, The third signaling message has the highest transmission priority.

11. The method according to claim 1, wherein when the first information includes a signaling message, the first information is periodically sent to the first GSP.

12. A fully scheduled Ethernet network comprising a control layer, a network layer, and a computing layer, wherein the network layer comprises a plurality of GSFs and a plurality of GSPs, wherein: For any first GSP among the multiple GSPs, the first GSP is used to determine a target uplink from multiple uplinks according to the first information and use the target uplink to send PKTC, and the multiple uplinks are links between the first GSP and the multiple GSFs.

13. An electronic device comprising: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method according to any one of claims 1 to 11. 14 . A computer-readable storage medium, wherein when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the method according to claim 1 .

Images