Data transmission device and method suitable for interconnection of multiple computing nodes

Abstract

The present disclosure relates to the field of communication technology, including a data transmission device and method suitable for multi-computing node interconnection. The device is connected with n optical transceiver modules respectively plugged with n computing nodes, a wavelength division multiplexer connected with the n optical transceiver modules, a splitter connected with the wavelength division multiplexer, n wavelength selectors connected with the splitter, and the n wavelength selectors are also connected with the n optical transceiver modules. Without advanced semiconductor process support, the device can solve the problems of complex implementation technology, difficult product manufacturing, expensive products, difficult wiring, and high latency in traditional multi-computing node interconnection solutions. The device improves data transmission speed and saves overall networking cost. The use of mature optical communication devices can reduce the manufacturing difficulty of the data transmission device. The optical channel uses wavelength as the scheduling unit, which can reduce the difficulty of rate upgrade, ignore device heterogeneity, and improve the flexibility of device application.

Description

Technical Field

[0001] This disclosure relates to the field of communication technology, and in particular to a data transmission device and method suitable for interconnecting multiple computing nodes. Background Technology

[0002] With the rapid development of artificial intelligence and machine learning technologies, the demand for computing power in electronic devices is constantly increasing. For example, training large models requires electronic devices with strong computing power. One way to improve the computing power of electronic devices is to enhance the computing power of each individual computing node within the device, but this approach has limited effectiveness. Another approach is to set up multiple computing nodes within the electronic device and interconnect them. This allows for massive data exchange and collaborative improvement of computing power among the multiple computing nodes. A computing node refers to a computing unit within an electronic device used to perform computational tasks, such as a Graphics Processing Unit (GPU) or a Central Processing Unit (CPU).

[0003] Currently, the methods for interconnecting various computing nodes include the following:

[0004] 1. Access between different computing nodes is achieved through the Peripheral Component Interconnect Express (PCIe) interface.

[0005] However, the PCIe interface has limitations in data transfer rate and bandwidth, which can lead to high communication latency between different computing nodes.

[0006] 2. A solution for interconnecting multiple computing nodes based on the NVLink protocol, namely NVSwitch. NVSwitch aims to solve the problem of efficient data transmission and computation in multi-computing node environments. NVLink bandwidth is 10 times that of PCIe. Furthermore, by adding more NVLink interfaces, NVSwitch can eliminate intermediate computing node jumps, expand the scale of computing node interconnection, and achieve uninterrupted operation of a fully interconnected computing node system, thereby enabling more powerful computing clusters. (Reference) Figure 1 The diagram shown illustrates the interconnection of multiple computing nodes based on NVSwitch. Figure 1 It is known that each NVSwitch and each computing node has multiple NVlink links, which are interconnected through point-to-point copper cable links.

[0007] However, the multi-node interconnection structure based on NVSwitch suffers from a large number of cables in the field. Furthermore, the NVSwitch chip requires 4nm manufacturing, which presents challenges in device manufacturing and high costs. Summary of the Invention

[0008] In view of this, this disclosure proposes a data transmission device and method suitable for interconnecting multiple computing nodes, which can solve the problems of complex implementation technology, difficult product manufacturing, expensive products, difficult wiring, and high latency in traditional interconnection schemes for multiple computing nodes.

[0009] According to one aspect of this disclosure, a data transmission device suitable for interconnecting multiple computing nodes is provided, the device comprising:

[0010] n optical transceiver modules are connected to n computing nodes respectively; each optical transceiver module is used to: convert the first service data to be sent by the corresponding computing node into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy when it receives the first service data to be sent by the corresponding computing node; where n is an integer greater than 1.

[0011] A wavelength division multiplexer connected to n optical transceiver modules is used to combine the first optical signals sent by each optical transceiver module to obtain a composite optical signal;

[0012] The splitter connected to the wavelength division multiplexer is used to transmit the composite optical signal to the wavelength selector corresponding to each computing node.

[0013] The n wavelength selectors connected to the splitter are also respectively plugged into n optical transceiver modules; each wavelength selector is used to: extract a second optical signal of a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters; and send the second optical signal to the corresponding optical transceiver module; wherein the wavelength selection parameters are adapted to the routing planning strategy;

[0014] The n optical transceiver modules are further configured to: upon receiving the second optical signal to be sent to the corresponding computing power node, convert the second optical signal into second service data and send it to the corresponding computing power node.

[0015] In one possible implementation, the optical transceiver module is further configured to acquire wavelength configuration parameters indicated by the routing planning strategy; wherein, the routing planning strategy includes the transmission mode of each type of service data and the mapping relationship between wavelength and transmission path; the service data includes the first service data and the second service data, and the wavelength configuration parameters include a first preset wavelength corresponding to the first service data; the transmission mode includes a single-wavelength transmission mode and a multi-wavelength transmission mode;

[0016] The wavelength selector is also used to obtain wavelength selection parameters that are compatible with the routing planning strategy; for the wavelength selector connected to the optical transceiver module of each computing node, the wavelength selection parameters of the wavelength selector are used to indicate the wavelength corresponding to the transmission path where the destination node is the computing node in the mapping relationship.

[0017] In one possible implementation, the apparatus further includes: a routing planner connected to each optical transceiver module and each computing node, the routing planner being used for:

[0018] Obtain the business requirements of the computing power nodes;

[0019] Based on the aforementioned business requirements, the routing planning strategy, as well as the wavelength configuration parameters and wavelength selection parameters adapted to the routing planning strategy, are determined.

[0020] The wavelength configuration parameters are sent to the optical transceiver module;

[0021] The wavelength selection parameters are sent to the wavelength selector.

[0022] In one possible implementation, the optical transceiver module is further configured to, in the event of an update to the service requirements of the computing power node, obtain wavelength configuration parameters of an updated routing planning strategy that are adapted to the service requirements, and execute, based on the updated wavelength configuration parameters, the step of converting the first service data into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy when the first service data to be sent by the corresponding computing power node is received, and subsequent steps.

[0023] Accordingly,

[0024] The wavelength selector is further configured to acquire updated wavelength selection parameters that are compatible with the updated routing planning strategy, and to use the updated wavelength selection parameters as the pre-configured wavelength selection parameters to execute the step of extracting a second optical signal with a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters, and subsequent steps.

[0025] In one possible implementation, the number n of computing nodes is determined based on the transmit optical power range of the optical transceiver module.

[0026] In one possible implementation, the apparatus further includes: a local switching module disposed within the computing power node, wherein the local switching module in each computing power node is used for:

[0027] Based on the service requirements of the computing power nodes, determine whether to forward data through the optical layer;

[0028] If it is determined that the forwarding is through the optical layer, the first service data of the computing power node is sent to the optical transceiver module corresponding to the computing power node, so as to trigger the optical transceiver module to execute the step of converting the first service data into a first optical signal of a first preset wavelength based on a preset routing planning strategy when the first service data to be sent by the corresponding computing power node is received.

[0029] In one possible implementation, before sending the first service data of the computing node to the optical transceiver module corresponding to the computing node, the local switching module is further configured to:

[0030] The data to be sent by the computing power node is divided into data segments to obtain the first service data;

[0031] Once it is determined that the data will be forwarded through the optical layer, each first service data will be sent to the optical transceiver module corresponding to the computing power node in a preset priority order.

[0032] In one possible implementation, the local switching module is further configured to:

[0033] If it is determined that data is to be forwarded through the electrical layer, the data to be sent by the computing node is forwarded to the local address indicated by the service requirement.

[0034] In one possible implementation, determining whether to forward data via the optical layer based on the service requirements of the computing node includes:

[0035] If the destination MAC address indicated by the service requirement does not belong to the local address table pre-stored by the local switching module, the data is forwarded through the optical layer.

[0036] According to another aspect of this disclosure, a data transmission method suitable for interconnecting multiple computing power nodes is provided, the method comprising:

[0037] Upon receiving the first service data to be sent from the corresponding computing power node, the first service data is converted into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy.

[0038] The first optical signals sent by each optical transceiver module are combined by a wavelength division multiplexer to obtain a composite optical signal.

[0039] The composite optical signal is transmitted to the wavelength selector corresponding to each computing node through a splitter.

[0040] The wavelength selector extracts a second optical signal with a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters; the second optical signal is then sent to the corresponding optical transceiver module; wherein the wavelength selection parameters are adapted to the routing planning strategy.

[0041] Upon receiving the second optical signal to be sent to the corresponding computing power node, the optical transceiver module converts the second optical signal into second service data and sends it to the corresponding computing power node.

[0042] By configuring n optical transceiver modules connected to n computing nodes, a wavelength division multiplexer (WDM) connected to each of the n WDM modules, a splitter connected to the WDM, and n wavelength selectors connected to the splitters, with the n wavelength selectors also connected to the n WDM modules, data transmission between different computing nodes is achieved through optical communication. Therefore, advanced semiconductor processes are not required, solving the problems of complex implementation, difficult manufacturing, high product cost, complex wiring, and high latency inherent in traditional multi-computing node interconnection schemes. This improves data transmission speed and reduces overall network costs. Furthermore, WDM and optical splitters are mature optical communication devices with relatively low technical requirements for manufacturing processes, thus reducing the manufacturing difficulty of data transmission devices. Additionally, the use of wavelength as the scheduling unit in the optical channel allows the optical communication system to easily handle smooth upgrades in future speeds, significantly reducing system upgrade costs. It also eliminates the need for device heterogeneity, enabling more flexible handling of various complex application scenarios.

[0043] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0044] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0045] Figure 1 This diagram illustrates the interconnection structure of multiple computing nodes based on NVSwitch.

[0046] Figure 2 This diagram illustrates a structural diagram of a data transmission apparatus suitable for interconnecting multiple computing nodes according to an embodiment of the present disclosure.

[0047] Figure 3 This diagram illustrates a data transmission apparatus based on wavelength division multiplexing (WDM) for interconnecting multiple computing nodes, according to an embodiment of the present disclosure.

[0048] Figure 4This diagram illustrates a computing node exchanging data based on a local switching module according to an embodiment of the present disclosure.

[0049] Figure 5 A flowchart illustrating a data transmission method applicable to the interconnection of multiple computing nodes according to an embodiment of the present disclosure is shown.

[0050] Figure 6 A flowchart illustrating a data transmission method applicable to the interconnection of multiple computing nodes according to another embodiment of this disclosure is shown. Detailed Implementation

[0051] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0052] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0053] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0054] Figure 2 This diagram illustrates a structural arrangement of a data transmission apparatus suitable for interconnecting multiple computing nodes according to an embodiment of the present disclosure. Figure 2 As shown, the device employs network-centric optical transmission technology, forming a two-tier architecture with multiple computing nodes. The network center introduces high-bandwidth, low-latency optical layer transmission technology; each computing node connects to other computing nodes through the network center. The computing nodes can be computing chips, computing boards, or computing racks, etc., and this application does not limit the implementation method of the computing nodes.

[0055] In this application, the network center based on optical transmission technology uses mature active and passive optical devices to realize optical layer networking.

[0056] Passive devices are used to achieve multiplexing and equal distribution of optical power. Passive devices are electronic devices that cannot actively amplify, adjust, or control signals. That is, "passive" means that the device has no active source of energy, but relies on the signal energy on the transmission line or the signal provided externally to achieve its function.

[0057] Active optical devices are used to achieve signal transmission and reception and wavelength selection. An active device is an electronic device that can actively amplify, adjust, or control signals. In other words, "active" refers to a source of energy or an active provider of signals.

[0058] The specific structure of the network center in this device is described below. According to... Figure 2 It is known that the device includes: an optical transceiver module 210, a wavelength division multiplexer 220, a splitter 230, and a wavelength selector 240.

[0059] The number of optical transceiver modules 210 is n, and each of the n optical transceiver modules is connected to one of the n computing nodes. Here, n is an integer greater than 1.

[0060] Each optical transceiver module 210 includes at least one optical transceiver, which has a bidirectional conversion function between optical signals and electrical signals, and uses a wavelength division multiplexing (WDM) mechanism to transmit and receive services. Exemplarily, each optical transceiver module 210 includes 1 to 8 optical transceivers. Each optical transceiver includes a transmitter (Tx) and a receiver (Rx), wherein the transmitter is used to convert the service data to be transmitted into an optical signal of a specified wavelength, and the Rx is used to convert the received optical signal into service data.

[0061] In this embodiment, the optical transceiver module 210 is not integrated into the computing node, but is plugged into the computing node as an independent module. This allows the data transmission device to be used for interconnection between different types of computing nodes, improving the versatility of the data transmission device. Furthermore, it eliminates the need to modify the computing node, reducing the manufacturing difficulty and cost of the data transmission device.

[0062] For example, when sending the first service data of each computing node, the optical transceiver module 210 connected to the computing node is used to: convert the first service data into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy when receiving the first service data to be sent by the corresponding computing node.

[0063] The first business data refers to the business data that the computing node is about to send.

[0064] In this embodiment, each computing node pre-stores a routing planning strategy corresponding to that computing node. This routing planning strategy is used to indicate the transmission method and optical transmission wavelength of service data with different service requirements in the computing node. For example, the routing planning strategy is a flow table rule stored in the computing node. Exemplarily, the routing planning strategy includes the transmission method of each type of service data and the mapping relationship between wavelength and transmission path. The service data includes the first service data mentioned above and the second service data mentioned below.

[0065] The routing planning strategy for each computing node is obtained through global planning based on information such as the topology and business requirements of each computing node, thereby ensuring that there are no conflicts between the routing planning strategies of different computing nodes. Optionally, the routing planning strategy can be determined manually, or it can be dynamically configured based on a routing planner, such as Software Defined Network (SDN). This embodiment does not limit the generation method of the routing planning strategy.

[0066] Transmission methods include single-wavelength transmission and multi-wavelength transmission. Single-wavelength transmission refers to a transmission method in which only a single wavelength of optical signal is used to transmit information in an optical communication system. Multi-wavelength transmission refers to a transmission method in which multiple different wavelengths of optical signal are used simultaneously to transmit information in an optical communication system.

[0067] Optionally, the routing planning strategy can be fixed, i.e., the routing planning strategy is static; or the routing planning strategy can be dynamically configured. This embodiment does not limit the implementation method of the routing planning strategy.

[0068] For example, the routing planning strategy for computing node 1 is shown in Table 1 below:

[0069] Table 1:

[0070] Business type Source node Target node Transmission method wavelength video stream Computing Node 1 Computing Node 2 single wavelength λ1 Data synchronization Computing Node 1 Computing Node 3 Multiwavelength λ2+λ3

[0071] According to Table 1, the routing planning strategy indicates that: for the first service data of the service type of computing node 1, which is video stream and is to be sent to computing node 2, it is converted into an optical signal with wavelength λ1 for single-wavelength transmission; for the first service data of the service type of computing node 1, which is data synchronization and is to be sent to computing node 3, it is converted into optical signals with wavelengths λ2 and λ3 for multi-wavelength transmission.

[0072] Optionally, the transmission wavelength of each optical transceiver in the optical transceiver module 210 is adjustable. In this case, the optical transceiver module 210 is also used to obtain wavelength configuration parameters indicated by the routing planning strategy. The wavelength configuration parameters indicate the transmission wavelength corresponding to each optical transceiver in the optical transceiver module 210, and these wavelength configuration parameters include at least a first preset wavelength corresponding to the first service data. Optionally, the wavelength configuration parameters of the optical transceiver module 210 can be sent by the computing node, or they can be sent by the routing planner; this embodiment does not limit the source of the wavelength configuration parameters.

[0073] In one example, when the computing node sends the first service data to the corresponding optical transceiver module 210, it can also send wavelength configuration parameters indicated by the routing planning strategy, such as λ1 or λ2, so that the optical transceiver module 210 can perform electro-optical conversion according to the wavelength configuration parameters.

[0074] In other implementations, the transmission wavelength of the optical transceiver in the optical transceiver module 210 can also be fixed, i.e., not adjustable. In this case, multiple optical transceivers can be set for each computing node according to the routing planning strategy. For example, computing node 1 can be set with three optical transceiver modules 210. The first optical transceiver module 210 is used to convert the electrical signal into an optical signal with wavelength λ1, the second optical transceiver module 210 is used to convert the electrical signal into an optical signal with wavelength λ2, and the third optical transceiver module 210 is used to convert the electrical signal into an optical signal with wavelength λ3.

[0075] In one possible implementation, if the routing strategy is determined by a routing planner, the data transmission apparatus provided in this embodiment further includes a routing planner connected to each optical transceiver module 210 and each computing node, respectively. The routing planner is used to: obtain the service requirements of the computing nodes; determine the routing strategy based on the service requirements, as well as wavelength configuration parameters and wavelength selection parameters adapted to the routing strategy; send the wavelength configuration parameters to the optical transceiver module 210; and send the wavelength selection parameters to the wavelength selector 240.

[0076] The service requirements are used to indicate the requirements for transmitting the first service data. For example, the service requirements include at least the address information of the source node and the destination node; the service requirements may also include the service type, bandwidth requirements, and / or latency constraints of the first service data, etc. This embodiment does not limit the content of the service requirements.

[0077] The routing planner determines the routing planning strategy based on business requirements in the following ways: obtaining global network status information and business requirements sent by computing nodes; and using a preset routing planning algorithm to allocate the optimal wavelength that meets the business requirements based on the global network status information.

[0078] The global network status information is used to indicate the available resources in the current optical network. This global network status information includes, but is not limited to, the list of currently available wavelengths, wavelength occupancy rate, and / or optical power budget. This embodiment does not limit the content of the global network status information.

[0079] Optionally, the routing planning algorithm can employ a hierarchical decision model, combining service requirements and physical constraints to achieve optimal wavelength allocation. For example, the hierarchical decision model selects low-noise, high-linearity wavelengths for high-bandwidth services (such as 4K video streams) and allocates multiple wavelengths to achieve multi-wavelength transmission; for low-latency services, it selects the wavelength corresponding to the shortest physical path. Alternatively, the routing planning algorithm can also be obtained by training a deep learning network using sample service requirements and the corresponding wavelength labels. This embodiment does not limit the implementation method of the routing planning algorithm.

[0080] The wavelength configuration parameters and wavelength selection parameters adapted to the routing planning strategy are as follows: For each group of wavelengths and transmission paths in the routing planning strategy, the wavelength indicated by the mapping relationship is determined as the wavelength configuration parameter corresponding to the source node and the wavelength selection parameter corresponding to the target node on the corresponding transmission path. For example, in Table 1 above, the mapping relationship includes the mapping relationship between wavelength λ1 and transmission path: from computing node 1 to computing node 2. At this time, when sending the first service data corresponding to the video stream, the wavelength configuration parameter of the optical transceiver module 210 plugged into computing node 1 is λ1, and the wavelength selection parameter of the wavelength selector 240 plugged into computing node 2 is λ1.

[0081] The wavelength division multiplexing (WDM) unit is connected to n optical transceiver modules 210 and is used to combine the first optical signals sent by each optical transceiver module 210 to obtain a composite optical signal.

[0082] For example, n optical transceiver modules 210 can multiplex the same wavelength division multiplexer 220. The wavelength division multiplexer 220 can use dense wavelength division multiplexing (DWDM) technology to aggregate the first service data output by multiple optical transceiver modules 210 from different computing nodes.

[0083] The splitter 230 is connected to the wavelength division multiplexer 220 and is used to transmit the composite optical signal to the wavelength selector 240 corresponding to each computing node.

[0084] For example, the splitter 230 realizes the service broadcast push between different computing power nodes through the passive optical splitting broadcast mechanism, thereby transmitting the composite optical signal to the wavelength selector 240 corresponding to each computing power node.

[0085] In this embodiment, the number of wavelength selectors 240 is also n. Each of the n wavelength selectors 240 is connected to a splitter 230 and also connected to one of the n optical transceiver modules 210. Specifically, the wavelength selector 240 is connected to the receiving end of the optical transceiver in the optical transceiver module 210 to send the selected optical signal to the receiving end, which then transmits it to the corresponding computing node. In this case, the n wavelength selectors 240 multiplex the same splitter 230. Optionally, the output power of the ports connected to the splitter 230 and each wavelength selector 240 is the same.

[0086] Each wavelength selector 240 is used to: extract a second optical signal with a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters; and send the second optical signal to the corresponding optical transceiver module 210; wherein the wavelength selection parameters are adapted to the routing planning strategy.

[0087] The wavelength selector 240's wavelength selection function can extract relevant service data for a single wavelength or multiple wavelengths required by the destination node from the entire optical network, thereby enabling internal interconnection and dynamic optical routing between any computing power nodes. For example: in Figure 2 In this system, computing node 1 is connected to computing node 3 through one wavelength channel, and computing node 2 is connected to computing node 1 through another wavelength channel.

[0088] Optionally, if the routing planning strategy can be dynamically configured, the wavelength selector 240 corresponding to different computing power nodes is also used to obtain wavelength selection parameters that are compatible with the routing planning strategy; for the wavelength selector 240 connected to the optical transceiver module 210 of each computing power node, the wavelength selection parameters of the wavelength selector 240 are used to indicate the wavelength corresponding to the transmission path where the destination node is the computing power node in the mapping relationship.

[0089] For example, in Table 1 above, for the wavelength selector 240 connected to the optical transceiver module 210 of computing node 3, the wavelength corresponding to the transmission path to the destination node is λ1, and the wavelength selection parameter of the wavelength selector 240 corresponding to computing node 3 is λ1.

[0090] The n optical transceiver modules 210 are also used to: upon receiving a second optical signal to be sent to the corresponding computing power node, convert the second optical signal into second service data and send it to the corresponding computing power node.

[0091] The second type of business data refers to the business data received by the computing power nodes.

[0092] In summary, the data transmission device for interconnecting multiple computing nodes provided in this embodiment comprises n optical transceiver modules connected to n computing nodes respectively, a wavelength division multiplexer connected to all n optical transceiver modules, a splitter connected to the wavelength division multiplexer, and n wavelength selectors connected to the splitter. The n wavelength selectors are also connected to the n optical transceiver modules. Since data transmission between different computing nodes is achieved through optical communication, advanced semiconductor processes are not required. This solves the problems of complex implementation technology, difficult product manufacturing, high product cost, complex wiring, and high latency inherent in traditional multi-computing node interconnection schemes. It can improve data transmission speed and save overall networking costs. Furthermore, wavelength division multiplexers and optical splitters are mature optical communication devices with relatively low technical requirements for manufacturing processes, thus reducing the manufacturing difficulty of the data transmission device.

[0093] In this embodiment, wavelength selection devices enable flexible wavelength scheduling, achieving dynamic optical routing between computing nodes. The optical channel uses wavelength as the scheduling unit, and its scheduling and allocation are independent of specific interface rates. Therefore, regardless of changes in the interface rate, as long as the wavelength of the optical signal remains consistent, correct information transmission can be ensured. Furthermore, due to the independence of wavelength scheduling, when communication rates need to be increased, there is no need for large-scale modifications or replacements of existing optical channels. Only corresponding adjustments are needed at the transmitting and receiving ends to ensure correct transmission and reception of signals at the new rate. This flexibility allows the optical communication system to easily handle smooth upgrades to future rates, significantly reducing system upgrade costs. For example, based on new rate requirements, it may be necessary to select higher-performance light sources, such as wavelength-tunable lasers or semiconductor lasers. Another example is adjusting the output power and wavelength of the light source to ensure matching with the receiving end and meeting new rate requirements. The photodetector at the receiving end is responsible for converting the received optical signal into an electrical signal. During rate upgrades, it may be necessary to select a photodetector with higher sensitivity and lower noise, and adjust the photodetector parameters, such as gain and bandwidth, to ensure accurate reception and conversion of high-speed optical signals.

[0094] Furthermore, as the physical carrier of information, wavelength does not depend on specific communication protocols during transmission. Therefore, when using wavelength as the scheduling unit in an optical channel, the specific implementation protocols within the computing nodes do not need to be considered. This protocol transparency allows optical communication systems to easily connect to various types of computing nodes without worrying about compatibility issues. In practical applications, computing nodes often have devices with different models, specifications, and performance. When using wavelength as the scheduling unit, the differences between these devices do not need to be considered; it is only necessary to ensure that they can send and receive optical signals of the specified wavelength. This disregard for device heterogeneity greatly simplifies the design and deployment process of optical communication systems. Because device heterogeneity can be ignored, optical communication systems can more flexibly cope with various complex application scenarios. For example, in scenarios such as cloud computing and big data, the number and type of computing nodes may change at any time. When using wavelength as the scheduling unit, no additional processing or adjustment is required for these changes, thereby improving the system's flexibility and scalability.

[0095] In addition, the network center adopts a pure optical passive networking scheme, which can reduce node failure rate and reduce photoelectric conversion latency.

[0096] Optionally, in the above embodiments, the example of n computing nodes and data transmission devices being implemented in different devices is used for illustration. In actual implementation, the n computing nodes can also be connected to the optical transceiver module 210 as part of the data transmission device. This embodiment does not limit this.

[0097] In one possible implementation, the number n of computing nodes connected to each network center is determined based on the transmit optical power range of the optical transceiver module.

[0098] The optical power at the receiver of the optical transceiver module is the transmitted optical power minus the total loss of the optical signal in the transmission path. The optical power at the receiver of the optical transceiver module needs to be higher than the receiving sensitivity (i.e., reliable transmission condition), otherwise the receiver cannot decode the signal correctly.

[0099] For example, in the entire optical transmission path, optical splitters, wavelength multiplexers, wavelength selectors, optical fibers, and optical fiber connectors are all part of the total loss of transmitted optical power. Among them, the insertion loss of optical splitters accounts for the largest proportion, and the insertion loss of optical splitters can be expressed by the following formula:

[0100] L = -10lg(split ratio) + connector insertion loss;

[0101] Where L represents the insertion loss of the optical splitter.

[0102] According to the above formula, the insertion loss of an optical splitter with 16 computing nodes is approximately 13 dB; with 32 computing nodes, it is approximately 16 dB; and with 64 computing nodes, it is approximately 20 dB. If the transmit optical power of the optical transceiver module is limited, the total loss corresponding to the number of computing nodes (n) must ensure that the above reliable transmission conditions are met. To increase the number of computing nodes in the entire system, the transmit power and receive sensitivity of the transceiver can be increased, or an insertion amplifier can be used.

[0103] Taking a computing node sending a 100G service as an example, the transmit optical power range of a 100G extended distance (ER) optical module is generally +0.5dBm to +4.5dBm, and the receive optical power range is generally -20.5dBm to -1.9dBm. Considering line losses due to WDM, connectors, etc., if the network is configured with a 1:32 optical splitting ratio, it can support the interconnection of 32 computing nodes. In this case, n = 32. Figure 3 As shown, each computing node connects to other computing nodes through different wavelengths, enabling point-to-point connections and / or interconnections of multiple wavelengths between one computing node and another, and / or multi-point or broadcast connections between one computing node and multiple computing nodes; the entire network uses general-purpose optical devices, and the technology is mature and stable.

[0104] In this embodiment, by determining the number n of computing nodes based on the range of transmitted optical power of the optical transceiver module, reliable data transmission between computing nodes can be guaranteed, thereby ensuring the transmission effect of the data transmission device.

[0105] Optionally, in the above embodiments, the routing planning strategy can be adaptively updated when the service requirements of the computing power node are updated. In this case, the optical transceiver module is further configured to obtain an updated routing planning strategy adapted to the service requirements when the service requirements of the computing power node are updated, and to use the updated routing planning strategy as a preset routing planning strategy to execute the step of converting the first service data to a first optical signal of a first preset wavelength based on the preset routing planning strategy, and subsequent steps.

[0106] Accordingly, the wavelength selector is also used to obtain updated wavelength selection parameters adapted to the updated routing planning strategy, so as to use the updated wavelength selection parameters as pre-configured wavelength selection parameters to perform the step of extracting a second optical signal with a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters and subsequent steps.

[0107] For example, as shown in Table 1 above, if the video stream service requires additional computing resources, the routing planning strategy will be updated to use multi-wavelength transmission, resulting in the updated routing planning strategy. Correspondingly, the wavelength configuration parameters for the optical transceiver module and the wavelength selection parameters for the wavelength selector need to be adaptively changed.

[0108] In this embodiment, by updating the routing planning strategy based on business needs and adaptively changing the wavelength configuration parameters corresponding to the optical transceiver module and the wavelength selection parameters corresponding to the wavelength selector, the data transmission device can transmit data in a way that meets business requirements, thereby improving the flexibility of the data transmission device.

[0109] Optionally, to achieve distributed local exchange and more granular packet-level service scheduling within the computing power node, the data transmission device in this embodiment further includes a local exchange module located within the computing power node. The local exchange module handles data exchange within the computing power node. For example: [reference needed] Figure 4 Each computing node includes at least two computing cores (e.g., a CPU and a GPU). A local switching module within this node enables data exchange between the CPU and GPU without requiring cross-node communication. For example, since the local switching module operates within the local computing node, and the number of computing cores within each node is limited, the local switching module can be designed as a small-port line-speed switch, thereby reducing the design cost of the computing node.

[0110] Accordingly, the local switching module in each computing node is used for:

[0111] Based on the service requirements of the computing power node, it is determined whether to forward data through the optical layer. If it is determined that data should be forwarded through the optical layer, the first service data of the computing power node is sent to the optical transceiver module corresponding to the computing power node, so as to trigger the optical transceiver module to execute the step of converting the first service data into a first optical signal of a first preset wavelength based on a preset routing planning strategy when it receives the first service data to be sent by the corresponding computing power node.

[0112] In one example, determining whether to forward data through the optical layer based on the service requirements of the computing power node includes: if the destination MAC address indicated by the service requirements does not belong to the local address table pre-stored by the local switching module, forwarding data through the optical layer, that is, sending the first service data of the computing power node to the optical transceiver module corresponding to the computing power node, so as to trigger the optical transceiver module to perform the step of converting the first service data into a first optical signal of a first preset wavelength based on a preset routing planning strategy when it receives the first service data to be sent by the corresponding computing power node.

[0113] Accordingly, if the destination MAC address indicated by the service request belongs to the local address table pre-stored by the local switching module, it is determined that the data to be sent by the computing node will be forwarded to the local address indicated by the service request via the electrical layer. In this case, the first service data is sent directly to the destination MAC address without needing to be forwarded via the optical layer.

[0114] Optionally, before sending the first service data of the computing node to the optical transceiver module corresponding to the computing node, the local switching module is further used for:

[0115] The data to be sent by the computing power node is divided into first service data; if it is determined that the forwarding will be carried out through the optical layer, each first service data is sent to the optical transceiver module corresponding to the computing power node in a preset priority order.

[0116] The methods for dividing the data to be sent include, but are not limited to: dividing according to protocol type and port number; and / or dividing according to the address information of the destination node, etc. This embodiment does not limit the method of dividing the data to be sent.

[0117] In this embodiment, the local switching module inside the computing node can realize distributed switching and service scheduling at the packet level with finer granularity; the multiple groups of first service data obtained have different priority orders, and each first service data is sent to the optical transceiver module corresponding to the computing node according to the preset priority order, which can reduce the latency of important services and improve the data transmission effect.

[0118] Figure 5 A flowchart illustrating a data transmission method applicable to the interconnection of multiple computing nodes according to an embodiment of this disclosure is shown. Figure 5As shown, the method includes the following steps:

[0119] Step 501: Upon receiving the first service data to be sent from the corresponding computing power node, the first service data is converted into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy.

[0120] Step 502: Combine the first optical signals sent by each optical transceiver module using a wavelength division multiplexer to obtain a composite optical signal;

[0121] Step 503: The composite optical signal is transmitted to the wavelength selector corresponding to each computing node through a splitter.

[0122] Step 504: Extract a second optical signal with a second preset wavelength from the received composite optical signal using a wavelength selector, according to pre-configured wavelength selection parameters; send the second optical signal to the corresponding optical transceiver module; wherein the wavelength selection parameters are adapted to the routing planning strategy.

[0123] Step 505: Upon receiving the second optical signal to be sent to the corresponding computing power node, the second optical signal is converted into second service data and sent to the corresponding computing power node through the optical transceiver module.

[0124] For relevant details, please refer to the above device embodiments, which will not be repeated here.

[0125] To better understand the data transmission process of the data transmission device provided in this application, the following example illustrates the process. Figure 6 , Figure 6 Taking the example of computing node 1 sending business data to computing node 2, the process includes the following steps:

[0126] Step 61: Computing node 1 obtains a routing planning strategy determined based on business requirements. The routing planning strategy includes the transmission method and the mapping relationship between wavelength and transmission path.

[0127] Step 62: The optical transceiver module of computing node 1 obtains the wavelength configuration parameters that match the routing planning strategy, and the wavelength selector corresponding to computing node 2 obtains the wavelength selection parameters that match the routing planning strategy.

[0128] Step 63: The local switching module in computing node 1 divides the data to be sent into data segments to obtain the first service data; and sends the first service data to optical transceiver module 1.

[0129] Step 64: The transmitting end of the optical transceiver module 1 converts the first service data into a first optical signal with a first preset wavelength according to the wavelength configuration parameters; and outputs the first optical signal to the wavelength division multiplexer through the optical interface.

[0130] Step 65: The wavelength division multiplexer combines the first optical signals output by each computing node into a composite optical signal and outputs the composite optical signal to the splitter.

[0131] Step 66: The splitter uses a power equalization mode to split one composite optical signal into multiple composite optical signals and broadcast them to the wavelength selectors corresponding to each computing node.

[0132] Step 67: The wavelength selector corresponding to computing node 2 selects a second optical signal with a second preset wavelength that matches the wavelength selection parameters, and sends the second optical signal to the optical transceiver module 2 plugged into computing node 2.

[0133] Step 68: The optical transceiver module 2 converts the second optical signal into an electrical signal to obtain the second service data, and sends the second service data to the computing node 2. The local switching module inside the computing node 2 can realize fine-grained electrical layer service scheduling and forwarding.

[0134] Optionally, if business requirements change, steps 61-62 can be re-executed to transmit business data according to the updated routing planning strategy.

[0135] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A data transmission device suitable for interconnecting multiple computing nodes, characterized in that, The device includes: n optical transceiver modules are connected to n computing nodes respectively; each optical transceiver module is used to: convert the first service data to be sent by the corresponding computing node into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy when it receives the first service data to be sent by the corresponding computing node; where n is an integer greater than 1. A wavelength division multiplexer connected to n optical transceiver modules is used to combine the first optical signals sent by each optical transceiver module to obtain a composite optical signal; The splitter connected to the wavelength division multiplexer is used to transmit the composite optical signal to the wavelength selector corresponding to each computing node. The n wavelength selectors connected to the splitter are also respectively plugged into n optical transceiver modules; each wavelength selector is used to: extract a second optical signal of a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters; and send the second optical signal to the corresponding optical transceiver module; wherein the wavelength selection parameters are adapted to the routing planning strategy; The n optical transceiver modules are further configured to: upon receiving the second optical signal to be sent to the corresponding computing power node, convert the second optical signal into second service data and send it to the corresponding computing power node.

2. The apparatus according to claim 1, characterized in that, The optical transceiver module is further configured to acquire wavelength configuration parameters indicated by the routing planning strategy; wherein, the routing planning strategy includes the transmission method of each type of service data and the mapping relationship between wavelength and transmission path; the service data includes the first service data and the second service data, and the wavelength configuration parameters include the first preset wavelength corresponding to the first service data; the transmission method includes single-wavelength transmission method and multi-wavelength transmission method; The wavelength selector is also used to obtain wavelength selection parameters that are compatible with the routing planning strategy; for the wavelength selector connected to the optical transceiver module of each computing node, the wavelength selection parameters of the wavelength selector are used to indicate the wavelength corresponding to the transmission path where the destination node is the computing node in the mapping relationship.

3. The apparatus according to claim 2, characterized in that, The device further includes: a routing planner connected to each optical transceiver module and each computing node, the routing planner being used for: Obtain the business requirements of the computing power nodes; Based on the aforementioned business requirements, the routing planning strategy, as well as the wavelength configuration parameters and wavelength selection parameters adapted to the routing planning strategy, are determined. The wavelength configuration parameters are sent to the optical transceiver module; The wavelength selection parameters are sent to the wavelength selector.

4. The apparatus according to claim 2, characterized in that, The optical transceiver module is further configured to, in the event of an update to the service requirements of the computing power node, obtain wavelength configuration parameters of an updated routing planning strategy that are adapted to the service requirements, and based on the updated wavelength configuration parameters, execute the step of converting the first service data into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy when the first service data to be sent by the corresponding computing power node is received, and subsequent steps. Accordingly, The wavelength selector is further configured to acquire updated wavelength selection parameters that are compatible with the updated routing planning strategy, and to use the updated wavelength selection parameters as the pre-configured wavelength selection parameters to execute the step of extracting a second optical signal with a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters, and subsequent steps.

5. The apparatus according to any one of claims 1 to 4, characterized in that, The number of computing nodes, n, is determined based on the range of transmitted optical power of the optical transceiver module.

6. The apparatus according to claim 1, characterized in that, The device further includes: a local switching module disposed within the computing power node, wherein the local switching module in each computing power node is used for: Based on the service requirements of the computing power nodes, determine whether to forward data through the optical layer; If it is determined that the forwarding is through the optical layer, the first service data of the computing power node is sent to the optical transceiver module corresponding to the computing power node, so as to trigger the optical transceiver module to execute the step of converting the first service data into a first optical signal of a first preset wavelength based on a preset routing planning strategy when the first service data to be sent by the corresponding computing power node is received.

7. The apparatus according to claim 6, characterized in that, Before sending the first service data of the computing power node to the optical transceiver module corresponding to the computing power node, the local switching module is further configured to: The data to be sent by the computing power node is divided into data segments to obtain the first service data; Once it is determined that the data will be forwarded through the optical layer, each first service data will be sent to the optical transceiver module corresponding to the computing power node in a preset priority order.

8. The apparatus according to claim 6, characterized in that, The local switching module is also used for: If it is determined that data is to be forwarded through the electrical layer, the data to be sent by the computing node is forwarded to the local address indicated by the service requirement.

9. The apparatus according to any one of claims 6 to 8, characterized in that, The determination of whether to forward data through the optical layer based on the service requirements of the computing power node includes: If the destination MAC address indicated by the service requirement does not belong to the local address table pre-stored by the local switching module, the data is forwarded through the optical layer.

10. A data transmission method suitable for interconnecting multiple computing power nodes, characterized in that, The method includes: Upon receiving the first service data to be sent from the corresponding computing power node, the first service data is converted into a first optical signal of a first preset wavelength indicated by a preset routing planning strategy. The first optical signals sent by each optical transceiver module are combined by a wavelength division multiplexer to obtain a composite optical signal. The composite optical signal is transmitted to the wavelength selector corresponding to each computing node through a splitter. The wavelength selector extracts a second optical signal with a second preset wavelength from the received composite optical signal according to the pre-configured wavelength selection parameters; the second optical signal is then sent to the corresponding optical transceiver module; wherein the wavelength selection parameters are adapted to the routing planning strategy. Upon receiving the second optical signal to be sent to the corresponding computing power node, the optical transceiver module converts the second optical signal into second service data and sends it to the corresponding computing power node.

Images