A communication system, method, device, medium and program product

Abstract

The application discloses a communication system, method, device, medium and program product in the computer technical field. In the application, any two GPUs in different servers need to pass through at most two switching modules to be interconnected, so that transmission delay of any two GPUs across servers is reduced; and a communication sensing module can realize communication link switching between any two GPUs across servers, efficient operation of an optoelectronic hybrid communication system and topology transformation across servers.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a communication system, method, device, medium and program product. Background Technology

[0002] In common hybrid optoelectronic switching systems, each GPU (Graphics Processing Unit) is directly connected to both the electrical and optical switching units. Furthermore, the bandwidth of the same GPU needs to be allocated between these two directly connected units, and the service types supported by each GPU must be configured. Because each GPU is directly connected to the same electrical and optical switching unit, the number of GPUs in the system is limited by the number of downstream ports of those units. Since the bandwidth of the same GPU is allocated between these units, when either the electrical or optical switching unit is not in operation, corresponding bandwidth resources are wasted.

[0003] Therefore, how to improve the communication efficiency of optoelectronic hybrid systems is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a communication system, method, device, medium, and program product to improve the communication efficiency of optoelectronic hybrid systems. The specific solution is as follows:

[0005] In a first aspect, this application provides a communication system, including: an optical switching module, a communication sensing module, and multiple servers;

[0006] The plurality of servers include: at least one switching module and a plurality of GPUs connected to the at least one switching module; the optical switching module and the communication sensing module are connected via an electrical communication link; the optical switching module and the communication sensing module are connected to at least one switching module in each server;

[0007] The communication sensing module is used to: determine, through a communication link with at least one switching module in each server, a first GPU and a second GPU that need to establish a communication connection, and send an optical path switching signal to the optical switching module via the electrical communication link to enable the first GPU and the second GPU to establish a communication connection; the first GPU and the second GPU are any two GPUs in different servers;

[0008] The optical switching module is used to: enable the first GPU and the second GPU to establish a communication connection according to the optical path switching signal.

[0009] Optionally, the communication sensing module is used to: obtain downstream port topology information of at least one switching module in each server from at least one switching module in each server through a communication link, and synchronize the downstream port topology information to at least one switching module in each server.

[0010] Optionally, the communication sensing module is used to: obtain downstream port topology information of at least one switching module in each server according to a scheduled task.

[0011] Optionally, the communication sensing module is used to: acquire in real time the downstream port topology information of at least one switching module in each server.

[0012] Optionally, the optical switching module is used to: control the transmission of optical signals between the optical waveguides of the switching modules to which the first GPU and the second GPU are respectively connected, according to the optical path switching signal.

[0013] Optionally, the communication sensing module is configured to: obtain a communication task flow from at least one switching module in each server through a communication link, and determine the first GPU and the second GPU according to the communication task flow.

[0014] Optionally, the communication sensing module is used to: construct the first GPU and the second GPU into a communication group; when there are multiple communication groups, arrange the multiple communication groups in sequence according to the communication task flow, and determine the communication duration of each communication group.

[0015] Optionally, the communication sensing module is used to: generate multiple optical path switching signals sequentially based on the communication duration of each communication group.

[0016] Optionally, the optical switching module includes: an optical waveguide network and at least one optical microring resonator;

[0017] The at least one optical micro-ring resonator is connected to the communication sensing module through the electrical communication link, and is used to control whether each optical waveguide in the optical waveguide network is conductive.

[0018] Each optical waveguide in the optical waveguide network is connected to at least one switching module in each server.

[0019] Optionally, the optical switching module and the communication sensing module are integrated into the same target device.

[0020] Optionally, there are multiple target devices; the optical switching module in any target device is connected to at least one switching module in each server through N first optical communication links; the communication sensing module in any target device is connected to at least one switching module in each server through N second optical communication links; N is the total number of switching modules.

[0021] Optionally, the communication sensing module in any target device is used to: when it confirms that it is in an idle state, perform the following steps: through the communication link between the device and at least one switching module in each server, determine the first GPU and the second GPU that need to establish a communication connection from at least one switching module in each server, and send an optical path switching signal to the optical switching module through the electrical communication link to enable the first GPU and the second GPU to establish a communication connection, and confirm that it is in a busy state during the execution process.

[0022] Optionally, the communication sensing module in any target device can be connected to a single switching module in any server via multiple second optical communication links.

[0023] Optionally, the communication sensing module in any target device is used to: sense the remaining bandwidth in the plurality of second optical communication links, select an idle second optical communication link in the plurality of second optical communication links according to the remaining bandwidth, and send the optical path switching signal through the idle second optical communication link.

[0024] Optionally, the communication sensing module in any target device is used to: select a second optical communication link with a bandwidth remaining greater than a preset threshold as an idle second optical communication link.

[0025] Optionally, the optical switching module is used to: cut off any optical communication link in the optical switching module that affects the establishment of the communication connection between the first GPU and the second GPU according to the optical path switching signal, and establish an optical communication link between the first GPU and the second GPU in the optical switching module.

[0026] Optionally, the communication sensing module is used to: read routing information stored in the corresponding switching module from at least one switching module in each server through a communication link, and determine the first GPU and the second GPU that need to establish a communication connection based on the routing information.

[0027] Secondly, this application provides a communication method applied to a communication sensing module, comprising:

[0028] By establishing a communication link with at least one switching module in each server, a first GPU and a second GPU that need to establish a communication connection are determined from at least one switching module in each server; wherein each server includes at least one switching module and multiple GPUs connected to the at least one switching module; the first GPU and the second GPU are any two GPUs from different servers;

[0029] An optical path switching signal is sent to the optical switching module via an electrical communication link connected to the optical switching module, enabling the optical switching module to establish a communication connection between the first GPU and the second GPU according to the optical path switching signal.

[0030] Optionally, a first GPU and a second GPU requiring a communication connection are determined from at least one switching module in each server via a communication link, including:

[0031] The communication task flow is obtained from at least one switching module of each server through the communication link between each server and at least one switching module of each server.

[0032] The first GPU and the second GPU are determined according to the communication task flow.

[0033] Thirdly, this application provides an electronic device, comprising:

[0034] Memory, used to store computer programs;

[0035] A processor for executing the computer program to implement the aforementioned disclosed communication method.

[0036] Fourthly, this application provides a non-volatile storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned disclosed communication method.

[0037] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the aforementioned disclosed communication method.

[0038] As can be seen from the above scheme, this application provides a communication system, including: an optical switching module, a communication sensing module, and multiple servers; wherein, the multiple servers include: at least one switching module and multiple GPUs connected to the at least one switching module; the optical switching module and the communication sensing module are connected via an electrical communication link; the optical switching module and the communication sensing module are connected to at least one switching module in each server; the communication sensing module is used to: determine, through the communication link with at least one switching module in each server, a first GPU and a second GPU that need to establish a communication connection from at least one switching module in each server, and send an optical path switching signal to the optical switching module to enable the first GPU and the second GPU to establish a communication connection via the electrical communication link; the first GPU and the second GPU are any two GPUs in different servers; the optical switching module is used to: enable the first GPU and the second GPU to establish a communication connection according to the optical path switching signal.

[0039] As can be seen, the beneficial effects of this application are as follows: any two GPUs in different servers can be interconnected through at most two switching modules, reducing the transmission latency between any two GPUs across servers; furthermore, the communication sensing module, through the communication link with at least one switching module in each server, determines any two GPUs in different servers that need to establish a communication connection from at least one switching module in each server, and sends the corresponding optical path switching signal to the optical switching module through the electrical communication link, so that the optical switching module establishes a communication connection between the two GPUs across servers according to the optical path switching signal. This realizes the switching of communication links between any two GPUs across servers, the efficient operation of the optoelectronic hybrid communication system, and the topology transformation across servers, thereby improving the communication efficiency of the optoelectronic hybrid system.

[0040] Correspondingly, the communication device, equipment, medium, and program product provided in this application also have the above-mentioned technical effects. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of a communication system disclosed in this application;

[0043] Figure 2 This is a schematic diagram of another communication system disclosed in this application;

[0044] Figure 3 for Figure 2 The diagram shows the workflow of the communication system.

[0045] Figure 4 for Figure 2 A schematic diagram of communication connection establishment in the communication system shown;

[0046] Figure 5 for Figure 2 A schematic diagram of topology switching in the communication system shown.

[0047] Figure 6 This is a schematic diagram of an optical switching switch disclosed in this application;

[0048] Figure 7 This is a flowchart illustrating the operation of an optical switching switch disclosed in this application.

[0049] Figure 8 This is a schematic diagram of the third communication system disclosed in this application;

[0050] Figure 9 This is a schematic diagram of the fourth communication system disclosed in this application;

[0051] Figure 10 This is a schematic diagram of an electronic device disclosed in this application;

[0052] Figure 11 A server architecture diagram provided in this application;

[0053] Figure 12 A terminal structure diagram provided for this application. Detailed Implementation

[0054] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other instances obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0055] Currently, each GPU is directly connected to both electrical and optical switching units. Furthermore, the bandwidth of the same GPU needs to be allocated between these two units, and the service types supported by each GPU must be configured. Because each GPU is directly connected to the same electrical and optical switching unit, the number of GPUs in the system is limited by the number of downstream ports of those units. Since the bandwidth of the same GPU is allocated between these units, bandwidth resources are wasted when either unit is not in operation. To address this, this application provides a communication scheme that enables communication link switching between any two GPUs across servers, efficient operation of the optoelectronic hybrid communication system, and topology transformation across servers, thereby improving the communication efficiency of the optoelectronic hybrid system.

[0056] See Figure 1 As shown in the figure, this application discloses a communication system, including: an optical switching module, a communication sensing module, and multiple servers.

[0057] The system comprises multiple servers, including at least one switching module and multiple GPUs connected to the at least one switching module. An optical switching module and a communication sensing module are connected via an electrical communication link. The optical switching module and the communication sensing module are connected to at least one switching module in each server. The switching module can be a PCIe switch, allowing more GPUs to be connected within the same server. PCI-express (Peripheral Component Interconnect express) is a high-speed serial computer expansion bus standard. The optical switching module is connected to at least one switching module in each server via an optical communication link, and the communication sensing module is connected to at least one switching module in each server via either an optical communication link or an electrical communication link.

[0058] The communication sensing module is used to: determine the first GPU and the second GPU that need to establish a communication connection from at least one switching module in each server through a communication link, and send an optical path switching signal to the optical switching module via an electrical communication link to enable the first GPU and the second GPU to establish a communication connection; the first GPU and the second GPU can be any two GPUs from different servers. The optical switching module is used to: enable the first GPU and the second GPU to establish a communication connection according to the optical path switching signal. The communication sensing module is equivalent to a controller, which controls the operation of the optical switching module through the optical path switching signal. The optical switching module is a switch with optical path switching function. It does not have a processor for data acquisition and data processing inside, that is, it does not have data sensing and data processing capabilities. Therefore, it needs to use the communication sensing module to sense routing information from at least one switching module in each server and generate a corresponding optical path switching signal to enable its operation.

[0059] According to this embodiment, the switching module is such as a PCIe switch. One switching module has multiple downstream ports, and each downstream port can connect to one GPU. Any two GPUs can be interconnected through a maximum of two PCIe switches, which can reduce transmission latency. Furthermore, this embodiment uses a communication sensing module to handle the topology sensing and topology switching of the entire system, enabling efficient switching operation of the optoelectronic hybrid communication system.

[0060] It's also important to note that each GPU connected to the PCIe switch chip has a corresponding address. When a GPU wants to communicate with other GPUs through the PCIe switch chip, it first prepares the data to be transmitted and sends it to a specific port of the connected PCIe switch chip according to the PCIe protocol format and specifications. This data contains the address information of the sending GPU and the address information of the receiving GPU (i.e., the destination GPU). After receiving this data, the PCIe switch chip, based on the destination address information, uses its internal routing algorithm and logic to determine which port the data should be forwarded from to reach the destination GPU. This routing information is stored in the internal memory of the PCIe switch. Therefore, the communication sensing module can read the internal memory of its connected PCIe switch through serial ports such as I2C or UART to obtain the routing information between the first and second GPUs that need to establish a communication connection. This routing information is used to identify the first and second GPUs. Therefore, in one embodiment, the communication sensing module is used to: read routing information stored in the corresponding switching module from the at least one switching module in each server through a communication link (specifically: read routing information stored in the memory inside the corresponding switching module), and determine the first GPU and the second GPU that need to establish a communication connection based on the routing information.

[0061] In one implementation, the communication awareness module's topology awareness of the entire system includes: the communication awareness module acquiring downstream port topology information of at least one switching module in each server from at least one switching module in each server through a communication link, and synchronizing the downstream port topology information to at least one switching module in each server. Thus, each server can obtain the system's global topology and the system's global GPU. Specifically, the communication awareness module is used to: acquire the downstream port topology information of at least one switching module in each server according to a scheduled task; or, acquire the downstream port topology information of at least one switching module in each server in real time.

[0062] In one embodiment, the optical switching module includes: an optical waveguide network and at least one optical microring resonator; the at least one optical microring resonator is connected to the communication sensing module via an electrical communication link, and is used to control whether each optical waveguide in the optical waveguide network is conductive; each optical waveguide in the optical waveguide network is connected to at least one switching module in each server. Accordingly, the optical switching module controls the optical waveguides between the switching modules to which the first GPU and the second GPU are respectively connected to transmit optical signals according to the optical path switching signal. An optical waveguide is a medium device that guides light waves to propagate therein, also known as a dielectric optical waveguide. In one example, the optical switching module is used to: cut off any optical communication link in the optical switching module that affects the establishment of the communication connection between the first GPU and the second GPU according to the optical path switching signal, and establish an optical communication link between the first GPU and the second GPU in the optical switching module. That is to say, the optical path switching signal can establish an optical communication link between the first GPU and the second GPU in the optical switching module, and this optical communication link is established after cutting off any optical communication link in the optical switching module that affects the establishment of the communication connection between the first GPU and the second GPU, thus realizing the optical path switching in the optical switching module. In this optical switching module, any optical communication link that affects the establishment of the communication connection between the first GPU and the second GPU can be: an optical communication link that is already connected to the first GPU or the second GPU, or an optical communication link that, although not connected to the first GPU or the second GPU, affects the establishment of the optical communication link between the first GPU and the second GPU. Therefore, in this embodiment, the optical path switching signal can be implemented in the optical switching module: to establish optical communication link A between the first GPU and the second GPU, any optical communication link affecting the establishment of optical communication link A is first cut off.

[0063] It should be noted that this embodiment allows the communication sensing module to establish communication links between any two GPUs across servers according to the communication order set in the communication task flow. This eliminates the need for the communication sensing module to perceive each switching module; communication links between any two GPUs across servers can be directly constructed sequentially, further improving communication efficiency. In one implementation, the communication sensing module is used to: obtain the communication task flow from at least one switching module in each server through the communication link, and determine the first GPU and the second GPU according to the communication task flow. The communication sensing module is also used to: construct the first GPU and the second GPU into a communication group; when there are multiple communication groups, arrange the multiple communication groups sequentially according to the communication task flow, and determine the communication duration of each communication group; multiple communication groups are multiple GPU groups, each group including GPUs from two different servers. In one example, the communication sensing module is used to: generate multiple optical path switching signals sequentially based on the communication duration of each communication group. Assuming there are three sequentially arranged communication groups according to the communication task flow: A (GPU1, GPU2), B (GPU1, GPU3), and C (GPU1, GPU4), with communication durations of 10 seconds, 100 seconds, and 60 seconds for groups A, B, and C, respectively, the communication sensing module generates a first optical path switching signal at 0 seconds, a second at 11 seconds, and a third at 111 seconds. These signals are immediately transmitted to the optical switching module after generation. During this process, the communication sensing module directly determines the first and second GPUs requiring a communication connection based on the communication task flow, eliminating the need for it to retrieve these GPUs from various servers. This reduces latency and improves communication efficiency. The time spent generating the optical path switching signals is negligible.

[0064] In one implementation, the optical switching module and the communication sensing module are integrated into the same target device. Furthermore, there can be multiple target devices in the system; the optical switching module in any target device is connected to at least one switching module in each server via N first optical communication links; the communication sensing module in any target device is connected to at least one switching module in each server via N second optical communication links; N is the total number of switching modules.

[0065] It should be noted that when there are multiple target devices in the system, the state of the communication sensing module in each target device can be set. When the communication sensing module is working, it is set to a busy state; when the communication sensing module is not working, it is set to an idle state. This allows for unified management of the communication sensing modules in each target device. When a new communication connection needs to be established, the communication sensing module in the idle state is selected by default to perform the following steps: through the communication link with at least one switching module in each server, the first GPU and the second GPU that need to establish a communication connection are determined from at least one switching module in each server, and an optical path switching signal that enables the first GPU and the second GPU to establish a communication connection is sent to the optical switching module in the current target device through an electrical communication link. Accordingly, the optical switching modules in each target device can also be managed uniformly. In one embodiment, the communication sensing module in any target device is used to: when it confirms that it is in an idle state, perform the step of determining the first GPU and the second GPU that need to establish a communication connection from at least one switching module in each server through the communication link with at least one switching module in each server, and sending an optical path switching signal that enables the first GPU and the second GPU to establish a communication connection is sent to the optical switching module through an electrical communication link, and confirm that it is in a busy state during the execution process.

[0066] To avoid competition among target devices for work, the following rule can be set: only one target device can establish and manage the same communication connection. To achieve this, a correspondence can be established between each communication connection and the communication sensing module and optical switching module in the corresponding target device. Specifically, when a communication sensing module is confirmed to be busy, a four-element correspondence is established between the current communication sensing module, the optical switching module belonging to the same target device as the current communication sensing module, and the first and second GPUs that the current communication sensing module determines need to establish a communication connection. If the communication connection is successfully established, the aforementioned four-element correspondence is released. Based on this four-element correspondence, it can be clearly determined which target device establishes each communication connection, avoiding the situation where different target devices compete to establish the same communication connection. For example, if any communication sensing module finds that the first and second GPUs needing to establish a communication connection already have a four-element correspondence, it will abandon the processing of the communication connection between the first and second GPUs.

[0067] In one implementation, a communication sensing module in any target device is connected to a single switching module in any server via multiple second optical communication links. That is, the single communication sensing module and the single switching module can have multiple optical communication links to achieve redundancy and alternatives between the two modules. In one example, the communication sensing module in the target device is used to: sense the bandwidth remaining amount among the multiple second optical communication links, select an idle second optical communication link based on the bandwidth remaining amount, and send an optical path switching signal through the idle second optical communication link. More specifically, the communication sensing module in the target device is used to: select a second optical communication link with a bandwidth remaining amount greater than a preset threshold as an idle second optical communication link.

[0068] As can be seen in this embodiment, any two GPUs in different servers can be interconnected through at most two switching modules, reducing the transmission latency between any two GPUs across servers. Furthermore, the communication sensing module, through the communication link with at least one switching module in each server, determines any two GPUs in different servers that need to establish a communication connection from at least one switching module in each server, and sends the corresponding optical path switching signal to the optical switching module through the electrical communication link, so that the optical switching module can establish a communication connection between the two GPUs across servers according to the optical path switching signal. This realizes the switching of communication links between any two GPUs across servers, the efficient operation of the optoelectronic hybrid communication system, and the topology transformation across servers, thereby improving the communication efficiency of the optoelectronic hybrid system.

[0069] Please see Figure 2 Another communication system configures multiple AI servers, each including a PCIe Switch chip (i.e., a switching module), with four GPUs connected to the PCIe Switch chip. Figure 2 In the diagram, 1 represents the AI ​​server, which internally contains a PCIe switch chip 11. This chip connects to multiple GPU cards 12 via a first electrical signal. 2 represents the optical communication sensing and switching module, which internally contains an optical switch 21 (i.e., the optical switching module) and a communication sensing module 22. The communication sensing module 22 connects to all servers via a second optical signal to obtain the GPU interconnect topology and communication switching requirements of the entire system. The communication sensing module 22 connects to the optical switch 21 via the second electrical signal to control the switching of its internal optical paths.

[0070] against Figure 2 For the system shown, please refer to the corresponding system workflow. Figure 3 Specifically, it includes:

[0071] S101: Topology acquisition: When the system is powered on and starts working, the communication sensing module 22 obtains the port topology of all internal PCIe Switch chips 11 of all servers 1 connected to the optical switching switch 21 through the second optical signal.

[0072] S102: Topology Distribution: The communication sensing module 22 distributes the obtained topology information to the internal PCIe Switch chip 11 of all servers 1 through the second optical signal.

[0073] S103: Topology awareness: When the GPU inside server 1 needs to switch topologies, the routing information reaches the PCIe Switch chip 11, and the communication awareness module 22 senses the topology switching requirement through the second optical signal.

[0074] S104: Topology switching: The communication sensing module 22 controls the optical switching switch 21 to change the internal optical path through the second electrical signal to realize the external interconnection topology switching of the PCIe Switch electrical chip 11.

[0075] Please see Figure 4 The initial communication topology operation process includes: GPU2 and GPU3, two of the four GPUs mounted on PCIe Switch chip 111, communicate through PCIe Switch chip 1. GPU1, after passing through routing path 101 in PCIe Switch chip 111, is converted into an optical signal, enters optical fiber 31, and then enters optical waveguide 41 in optical switch 21. It then reaches PCIe Switch chip 112 via optical waveguide 41 and establishes communication with GPU5 via routing path 104. Similarly, GPU4, after passing through routing path 103 in PCIe Switch chip 111, enters PCIe Switch chip 112 via optical fiber 31 and optical waveguide 41, and then establishes communication with GPU8 via routing path 105. At this time, the GPU in PCIe Switch chip 103 does not yet have a communication requirement with the GPUs mounted on other PCIe Switch chips.

[0076] Please see Figure 5The topology switching process includes: switching from communication between GPU4 and GPU8 to communication between GPU4 and GPU12, and establishing communication between GPU7 and GPU9. The switching is completed through the internal signal paths of PCIe Switch chips 111 and 112 and optical switch 21. GPU4 enters PCIe Switch chip 113 via routing path 103 in PCIe Switch chip 111, then via optical fiber 31 and optical waveguide 41, and then via routing path 108 to establish communication with GPU12. GPU7 enters PCIe Switch chip 113 via routing path 106 in PCIe Switch chip 111, then via optical fiber 31 and optical waveguide 41, and then via routing path 107 to establish communication with GPU9.

[0077] Please see Figure 6 The optical switching switch 21 mainly includes an optical waveguide 41 for transmitting PCIe optical signals and optical microring resonators 42, 43, and 44 for changing the propagation path of the optical signals in the optical waveguide 41. The optical waveguide 41 is connected to different PCIe switch electrical chips through optical fibers 31, 32, and 33, and the optical microring resonators 42, 43, and 44 are controlled by the communication sensing module 22 through a second electrical signal 23.

[0078] Please see Figure 7 The workflow corresponding to the optical switching switch may include:

[0079] S201: Optical path switching switch 21 is interconnected with the PCIe Switch chip in the server via optical fibers 31, 32 and 33. Optical fibers 31, 32 and 33 are interconnected with the optical waveguide network 41 inside optical path switching switch 21.

[0080] S202: When the optical micro-ring resonators 42 and 43 are working, the PCIe optical signal 34 entering through the optical fiber 31 enters the optical waveguide network 41 and is then transmitted out through the optical fiber 32, thereby realizing the communication between the GPU card mounted on the PCIe Switch chip in the server connected by the optical fibers 31 and 32.

[0081] S203: When the communication sensing module 22 receives the GPU communication switching request in the server connected by the optical fiber 31 through the second optical signal 24, it needs to establish communication with the GPU in the server connected by the optical fiber 33 to perform data interaction.

[0082] S204: The communication sensing module 22 controls the micro-ring resonators 42 and 44 in the optical path switching switch 21 to work and turn off 43 through the second electrical signal 23, so that the PCIe optical signal 34 entering through the optical fiber 31 enters the optical waveguide network 41 and is transmitted out through the optical fiber 33.

[0083] S205: Establish communication between the GPU mounted on the PCIe Switch chip of the server connected by fiber optic cable 31 and the GPU mounted on the PCIe Switch chip of the server connected by fiber optic cable 33.

[0084] Furthermore, in S201, it needs to be clarified that a single PCIe Switch chip has multiple optical fibers connected to the optical switching switch. When switching optical paths, it will not affect the communication between other GPUs under the same PCIe Switch chip and other GPUs.

[0085] Furthermore, in S203, in addition to controlling the optical switching switch to switch the optical path after the GPU generates a communication switching requirement, that is, after receiving a clear switching signal, the communication sensing module 22 can also control the optical switching switch to switch the optical path in advance according to the workflow of the communication library to which the GPU belongs in the AI ​​cluster, that is, to establish a communication connection according to the communication task flow, so as to further shorten the communication link delay.

[0086] Please see Figure 8 Two optical communication links can be set between an optical communication sensing switching module and a server. Figure 8 The system shown includes an optical communication sensing switching module 2 and ten servers: server 1, ..., server 10. The external ports of the PCIe switch chips in all servers are connected to the optical communication sensing switching module 2 via optical fiber.

[0087] Please see Figure 9 Multiple optical communication sensing switching modules can be set up in the same system. Figure 9 The system shown includes multiple optical communication sensing switching modules: optical communication sensing switching module 2, ..., optical communication sensing switching module 20; and ten servers: server 1, ..., server 10. The n external ports of the PCIe switch chips in the servers are connected to the n optical communication sensing switching modules via n optical fibers. Each optical communication sensing switching module has an optical fiber communication channel with all the PCIe chips.

[0088] This embodiment utilizes a communication sensing module to handle topology acquisition, distribution, sensing, and switching for the entire cluster, achieving efficient switching operation of the optoelectronic hybrid interconnect architecture. The communication sensing module also accesses the GPU communication library, enabling optical path pre-switching and reducing transmission latency. Based on PCIe switch electrical chips and PCIe optical interconnect technology, and leveraging optical switches to address the limited channel count of PCIe switch electrical chips, the number of GPUs interconnected via the PCIe protocol can be significantly expanded. Furthermore, thanks to the optoelectronic hybrid interconnect architecture, any two GPUs within the constructed cluster can be interconnected via a maximum of two PCIe switch electrical chips, avoiding the high latency issues associated with multi-level PCIe switch electrical chip networking. The PCIe optical interconnect technology extends the transmission distance of PCIe signals, thus solving the PCIe electrical interconnect loss problem.

[0089] The following describes a communication method provided by an embodiment of this application. The communication method described below can be referred to in conjunction with other embodiments described herein.

[0090] This application discloses a communication method applied to a communication sensing module, including:

[0091] S1. Through the communication link with at least one switching module in each server, determine the first GPU and the second GPU from at least one switching module in each server that need to establish a communication connection.

[0092] Each server includes at least one switching module and multiple GPUs connected to the at least one switching module; the first GPU and the second GPU are any two GPUs from different servers. In one example, a communication system may simultaneously include an optical switching module, a communication sensing module, and multiple servers; the optical switching module and the communication sensing module are electrically connected; the optical switching module and the multiple servers, and the communication sensing module and the multiple servers, are all optical path connections.

[0093] S2. Through the electrical communication link connected to the optical switching module, send an optical path switching signal to the optical switching module to enable the first GPU and the second GPU to establish a communication connection, so that the optical switching module enables the first GPU and the second GPU to establish a communication connection according to the optical path switching signal.

[0094] In one implementation, determining a first GPU and a second GPU that need to establish a communication connection from at least one switching module in each server via a communication link includes: obtaining a communication task flow from at least one switching module in each server via a communication link; and determining the first GPU and the second GPU according to the communication task flow.

[0095] In one embodiment, the communication sensing module is used to: obtain downstream port topology information of at least one switching module in each server from at least one switching module in each server through a communication link, and synchronize the downstream port topology information to at least one switching module in each server.

[0096] In one implementation, the communication sensing module is used to: acquire downstream port topology information of at least one switching module in each server according to a scheduled task.

[0097] In one implementation, the communication sensing module is used to: acquire downstream port topology information of at least one switching module in each server in real time.

[0098] In one embodiment, the optical switching module is used to: control the transmission of optical signals between the optical waveguides of the switching modules to which the first GPU and the second GPU are respectively connected, according to the optical path switching signal.

[0099] In one implementation, the communication sensing module is used to: obtain a communication task flow from at least one switching module in each server through a communication link, and determine a first GPU and a second GPU according to the communication task flow.

[0100] In one implementation, the communication sensing module is used to: construct a communication group for the first GPU and the second GPU; when there are multiple communication groups, arrange the multiple communication groups in sequence according to the communication task flow, and determine the communication duration of each communication group.

[0101] In one implementation, the communication sensing module is used to: generate multiple optical path switching signals sequentially based on the communication duration of each communication group.

[0102] In one embodiment, the optical switching module includes: an optical waveguide network and at least one optical microring resonator;

[0103] At least one optical micro-ring resonator is connected to the communication sensing module via an electrical communication link, and is used to control whether each optical waveguide in the optical waveguide network is conductive;

[0104] Each optical waveguide in the optical waveguide network is connected to at least one switching module in each server.

[0105] In one implementation, the optical switching module and the communication sensing module are integrated into the same target device.

[0106] In one implementation, there are multiple target devices; the optical switching module in any target device is connected to at least one switching module in each server through N first optical communication links; the communication sensing module in any target device is connected to at least one switching module in each server through N second optical communication links; N is the total number of switching modules.

[0107] In one embodiment, the communication sensing module in any target device is used to: when it is confirmed to be in an idle state, perform the following steps: through the communication link between the device and at least one switching module in each server, determine the first GPU and the second GPU that need to establish a communication connection from at least one switching module in each server, and send an optical path switching signal to the optical switching module through the electrical communication link to enable the first GPU and the second GPU to establish a communication connection, and confirm that it is in a busy state during the execution process.

[0108] In one implementation, the communication sensing module in any target device is connected to a single switching module in any server via multiple second optical communication links.

[0109] In one embodiment, the communication sensing module in any target device is used to: sense the bandwidth remaining amount in multiple second optical communication links, select an idle second optical communication link from among the multiple second optical communication links according to the bandwidth remaining amount, and send an optical path switching signal through the idle second optical communication link.

[0110] In one implementation, the communication sensing module in any target device is used to: select a second optical communication link with a bandwidth remaining greater than a preset threshold as an idle second optical communication link.

[0111] For more detailed information on the working process of each module and unit in this embodiment, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.

[0112] As can be seen, this embodiment provides a communication method that enables communication link switching between any two GPUs across servers, efficient operation of the optoelectronic hybrid communication system, and topology transformation across servers, thereby improving the communication efficiency of the optoelectronic hybrid system.

[0113] The following describes an electronic device provided by an embodiment of this application. The electronic device described below can be referred to in conjunction with other embodiments described herein. This electronic device may be a communication sensing module, an optical switching module, and other functional modules or devices described in the foregoing embodiments.

[0114] See Figure 10 As shown in the figure, an embodiment of this application discloses an electronic device, including:

[0115] Memory 1001 is used to store computer programs;

[0116] The processor 1002 is configured to execute the computer program to implement the method disclosed in any of the above embodiments.

[0117] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: through the communication link with at least one switching module in each server, determine the first GPU and the second GPU that need to establish a communication connection from at least one switching module in each server; through the electrical communication link connected to the optical switching module, send an optical path switching signal to the optical switching module to enable the first GPU and the second GPU to establish a communication connection, so that the optical switching module enables the first GPU and the second GPU to establish a communication connection according to the optical path switching signal.

[0118] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: obtaining the communication task flow from at least one switching module in each server through the communication link; and determining the first GPU and the second GPU according to the communication task flow.

[0119] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: through the communication link between the processor and at least one switching module in each server, obtain the downstream port topology information of at least one switching module in each server from at least one switching module in each server, and synchronize the downstream port topology information to at least one switching module in each server.

[0120] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: obtain the downstream port topology information of at least one switching module in each server according to a scheduled task.

[0121] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: real-time acquisition of the downstream port topology information of at least one switching module in each server.

[0122] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: according to the optical path switching signal, control the optical waveguides between the switching modules connected to the first GPU and the second GPU respectively to transmit optical signals.

[0123] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: obtain the communication task flow from at least one switching module in each server through the communication link, and determine the first GPU and the second GPU according to the communication task flow.

[0124] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: construct the first GPU and the second GPU into a communication group; when there are multiple communication groups, arrange the multiple communication groups in sequence according to the communication task flow, and determine the communication duration of each communication group.

[0125] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: generating multiple optical path switching signals in sequence based on the communication duration of each communication group.

[0126] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: connecting at least one switching module in each server through N first optical communication links; the communication sensing module in any target device connecting at least one switching module in each server through N second optical communication links; where N is the total number of switching modules.

[0127] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: when it confirms that it is in an idle state, it executes the step of determining the first GPU and the second GPU that need to establish a communication connection from the first GPU and the second GPU through the communication link between the processor and the at least one switching module in each server, and sending the optical path switching signal that enables the first GPU and the second GPU to establish a communication connection to the optical switching module through the electrical communication link, and confirming that it is in a busy state during the execution process.

[0128] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: sensing the bandwidth remaining amount in multiple second optical communication links, selecting an idle second optical communication link from among the multiple second optical communication links according to the bandwidth remaining amount, and sending an optical path switching signal through the idle second optical communication link.

[0129] In this embodiment, when the processor executes the computer program stored in the memory, it can specifically implement the following steps: selecting a second optical communication link with a bandwidth remaining greater than a preset threshold as an idle second optical communication link.

[0130] Furthermore, embodiments of this application also provide an electronic device. The aforementioned electronic device can be, for example,... Figure 11 The server shown can also be as follows: Figure 12 The terminal shown. Figure 11 and Figure 12 These are all diagrams illustrating the structure of an electronic device according to an exemplary embodiment. The content in the diagrams should not be considered as any limitation on the scope of this application.

[0131] Figure 11 This is a schematic diagram of a server structure provided in an embodiment of this application. The server may specifically include: at least one processor, at least one memory, a power supply, a communication interface, an input / output interface, and a communication bus. The memory stores a computer program, which is loaded and executed by the processor to implement the relevant steps in the communication disclosed in any of the foregoing embodiments.

[0132] In this embodiment, the power supply is used to provide operating voltage for each hardware device on the server; the communication interface can create a data transmission channel between the server and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0133] In addition, the memory, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system, computer programs and data, etc., and the storage method can be temporary storage or permanent storage.

[0134] The operating system manages and controls the various hardware devices and computer programs on the server to enable the processor to perform operations and processes on the data in the memory. It can be Windows Server, Netware, Unix, Linux, etc. In addition to computer programs capable of performing the communication methods disclosed in any of the foregoing embodiments, the computer programs may further include computer programs capable of performing other specific tasks. The data may include application update information and application developer information.

[0135] Figure 12This is a schematic diagram of the structure of a terminal provided in an embodiment of this application. The terminal may include, but is not limited to, a smartphone, tablet computer, laptop computer, or desktop computer.

[0136] Typically, the terminal in this embodiment includes a processor and a memory.

[0137] The processor may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor can be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor may also include a main processor and coprocessors. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor may also include an AI (Artificial Intelligence) processor, which handles computational operations related to machine learning.

[0138] The memory may include one or more computer non-volatile storage media, which may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory is used to store at least the following computer programs, which, after being loaded and executed by the processor, are capable of implementing the relevant steps in the communication method executed by the terminal side as disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory may also include operating systems and data, and the storage method may be temporary or permanent storage. The operating system may include Windows, Unix, Linux, etc. The data may include, but is not limited to, application update information.

[0139] In some embodiments, the terminal may further include a display screen, an input / output interface, a communication interface, a sensor, a power supply, and a communication bus.

[0140] Those skilled in the art will understand that Figure 12The structure shown does not constitute a limitation on the terminal and may include more or fewer components than illustrated.

[0141] The following describes a non-volatile storage medium provided in an embodiment of this application. The non-volatile storage medium described below can be referred to in conjunction with other embodiments described herein.

[0142] A non-volatile storage medium is provided for storing a computer program, wherein the computer program, when executed by a processor, implements the communication method disclosed in the foregoing embodiments. The non-volatile storage medium is a computer-readable non-volatile storage medium, which, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored thereon include an operating system, computer programs, and data, and the storage method can be temporary storage or permanent storage.

[0143] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: through a communication link with at least one switching module in each server, determine the first GPU and the second GPU that need to establish a communication connection from at least one switching module in each server; through an electrical communication link connected to the optical switching module, send an optical path switching signal to the optical switching module to enable the first GPU and the second GPU to establish a communication connection, so that the optical switching module enables the first GPU and the second GPU to establish a communication connection according to the optical path switching signal.

[0144] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: obtaining the communication task flow from at least one switching module in each server through the communication link; and determining the first GPU and the second GPU according to the communication task flow.

[0145] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: through the communication link between the processor and at least one switching module in each server, obtain the downstream port topology information of at least one switching module in each server from at least one switching module in each server, and synchronize the downstream port topology information to at least one switching module in each server.

[0146] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: obtaining the downstream port topology information of at least one switching module in each server according to a scheduled task.

[0147] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: real-time acquisition of the downstream port topology information of at least one switching module in each server.

[0148] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: according to the optical path switching signal, control the optical waveguides between the switching modules connected to the first GPU and the second GPU to transmit optical signals.

[0149] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: obtain the communication task flow from at least one switching module in each server through the communication link, and determine the first GPU and the second GPU according to the communication task flow.

[0150] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: construct the first GPU and the second GPU into a communication group; when there are multiple communication groups, arrange the multiple communication groups in sequence according to the communication task flow, and determine the communication duration of each communication group.

[0151] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: generating multiple optical path switching signals in sequence based on the communication duration of each communication group.

[0152] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: connecting at least one switching module in each server through N first optical communication links; the communication sensing module in any target device connecting at least one switching module in each server through N second optical communication links; N is the total number of switching modules.

[0153] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: when it confirms that it is in an idle state, it executes the step of determining the first GPU and the second GPU that need to establish a communication connection from the first GPU and the second GPU through the communication link with at least one switching module in each server, and sending the optical path switching signal that enables the first GPU and the second GPU to establish a communication connection to the optical switching module through the electrical communication link, and confirming that it is in a busy state during the execution process.

[0154] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: sensing the bandwidth remaining amount in multiple second optical communication links, selecting an idle second optical communication link from among the multiple second optical communication links according to the bandwidth remaining amount, and sending an optical path switching signal through the idle second optical communication link.

[0155] In this embodiment, when the processor executes the computer program stored in the non-volatile storage medium, it can specifically implement the following steps: selecting a second optical communication link with a bandwidth remaining greater than a preset threshold as an idle second optical communication link.

[0156] The following describes a computer program product provided by an embodiment of this application. The computer program product described below can be referred to in conjunction with other embodiments described herein.

[0157] A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the aforementioned disclosed communication method.

[0158] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0159] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of non-volatile storage medium known in the art.

[0160] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A communication system, characterized in that, include: Optical switching module, communication sensing module, and multiple servers; The plurality of servers include: at least one switching module and a plurality of GPUs connected to the at least one switching module; the optical switching module and the communication sensing module are connected via an electrical communication link; the optical switching module and the communication sensing module are connected to at least one switching module in each server; The communication sensing module is used to: determine, through a communication link with at least one switching module in each server, a first GPU and a second GPU that need to establish a communication connection, and send an optical path switching signal to the optical switching module via the electrical communication link to enable the first GPU and the second GPU to establish a communication connection; the first GPU and the second GPU are any two GPUs in different servers; The optical switching module is used to: enable the first GPU and the second GPU to establish a communication connection according to the optical path switching signal.

2. The communication system according to claim 1, characterized in that, The communication sensing module is used to: obtain downstream port topology information of at least one switching module in each server from at least one switching module in each server through a communication link, and synchronize the downstream port topology information to at least one switching module in each server.

3. The communication system according to claim 2, characterized in that, The communication sensing module is used to: obtain the downstream port topology information of at least one switching module in each server according to a scheduled task.

4. The communication system according to claim 2, characterized in that, The communication sensing module is used to: acquire the downstream port topology information of at least one switching module in each server in real time.

5. The communication system according to claim 1, characterized in that, The optical switching module is used to control the transmission of optical signals between the optical waveguides of the switching modules to which the first GPU and the second GPU are respectively connected, according to the optical path switching signal.

6. The communication system according to claim 1, characterized in that, The communication sensing module is used to: obtain the communication task flow from at least one switching module in each server through the communication link, and determine the first GPU and the second GPU according to the communication task flow.

7. The communication system according to claim 6, characterized in that, The communication sensing module is used to: construct the first GPU and the second GPU into a communication group; when there are multiple communication groups, arrange the multiple communication groups in sequence according to the communication task flow, and determine the communication duration of each communication group.

8. The communication system according to claim 7, characterized in that, The communication sensing module is used to generate multiple optical path switching signals in sequence based on the communication duration of each communication group.

9. The communication system according to claim 1, characterized in that, The optical switching module includes: an optical waveguide network and at least one optical microring resonator; The at least one optical micro-ring resonator is connected to the communication sensing module through the electrical communication link, and is used to control whether each optical waveguide in the optical waveguide network is conductive. Each optical waveguide in the optical waveguide network is connected to at least one switching module in each server.

10. The communication system according to any one of claims 1 to 9, characterized in that, The optical switching module and the communication sensing module are integrated into the same target device.

11. The communication system according to claim 10, characterized in that, There are multiple target devices; the optical switching module in any target device is connected to at least one switching module in each server through N first optical communication links; the communication sensing module in any target device is connected to at least one switching module in each server through N second optical communication links; N is the total number of switching modules.

12. The communication system according to claim 11, characterized in that, The communication sensing module in any target device is used to: when it confirms that it is in an idle state, perform the following steps: through the communication link between the device and at least one switching module in each server, determine the first GPU and the second GPU that need to establish a communication connection from at least one switching module in each server, and send an optical path switching signal to the optical switching module through the electrical communication link to enable the first GPU and the second GPU to establish a communication connection, and confirm that it is in a busy state during the execution process.

13. The communication system according to claim 11, characterized in that, The communication sensing module in any target device is connected to a single switching module in any server through multiple second optical communication links.

14. The communication system according to claim 13, characterized in that, The communication sensing module in any target device is used to: sense the remaining bandwidth in the plurality of second optical communication links, select an idle second optical communication link in the plurality of second optical communication links according to the remaining bandwidth, and send the optical path switching signal through the idle second optical communication link.

15. The communication system according to claim 14, characterized in that, The communication sensing module in any target device is used to: select a second optical communication link with a bandwidth remaining amount greater than a preset threshold as an idle second optical communication link.

16. The communication system according to any one of claims 1 to 9, characterized in that, The optical switching module is used to: cut off any optical communication link in the optical switching module that affects the establishment of the communication connection between the first GPU and the second GPU according to the optical path switching signal, and establish an optical communication link between the first GPU and the second GPU in the optical switching module.

17. The communication system according to any one of claims 1 to 9, characterized in that, The communication sensing module is used to: read routing information stored in the corresponding switching module from at least one switching module in each server through a communication link, and determine the first GPU and the second GPU that need to establish a communication connection based on the routing information.

18. A communication method, characterized in that, Applied to communication sensing modules, including: By establishing a communication link with at least one switching module in each server, a first GPU and a second GPU that need to establish a communication connection are determined from at least one switching module in each server; wherein each server includes at least one switching module and multiple GPUs connected to the at least one switching module; the first GPU and the second GPU are any two GPUs from different servers; An optical path switching signal is sent to the optical switching module via an electrical communication link connected to the optical switching module, enabling the optical switching module to establish a communication connection between the first GPU and the second GPU according to the optical path switching signal.

19. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in claim 18.

20. A non-volatile storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the method as described in claim 18.

21. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the method as described in claim 18.

Images