Task processing method and device based on photoelectric cooperation and computer equipment

Abstract

The invention relates to a task processing method and device based on photoelectric cooperation and computer equipment. The method comprises the following steps: receiving a task request, and determining a target data packet corresponding to the task request; performing feature matching based on the task label of the target data packet in each service chain template contained in a preset service chain template library, and determining a target service chain template corresponding to the target data packet; based on the service chain identifier and the task label of the target service chain template, packaging the target data packet to obtain a packaged data packet and a photoelectric resource association table; and transmitting the packaged data packet to the target terminal through the photoelectric resource association table, the optical switch, the electric switch and the transmission path, so that the target terminal performs task processing based on the packaged data packet to obtain a task result. Through the method, accurate binding of the computing power demand of the data packet and the photoelectric network resource is realized based on the task label of the data packet, and the utilization rate of the photoelectric network resource is improved under the condition that the data transmission performance is ensured.

Description

Technical Field

[0001] This application relates to the field of wide area network technology, and in particular to a task processing method, apparatus and computer device based on optoelectronic coordination. Background Technology

[0002] As intelligent computing clusters evolve from single-node kilocard / ten-thousand-card configurations to multi-node massive card interconnections, electrical switching networking architectures, limited by bandwidth, latency, energy consumption, scalability, and cost, cannot meet the ever-growing scale of intelligent computing clusters. Optical switching, with its advantages of high bandwidth, low latency, and high energy efficiency, can meet the needs of these increasingly large-scale intelligent computing clusters.

[0003] However, the intelligent computing interconnection of wide-area optoelectronic switching collaboration involves interconnecting multiple nodes in a wide area network, which involves multiple nodes performing different tasks. When allocating resources to each node, the related technologies are based on a relatively simple association method, which fixedly allocates network resources to each node, resulting in low resource utilization of the wide area network. Summary of the Invention

[0004] Therefore, it is necessary to provide a task processing method, apparatus, and computer equipment based on optoelectronic collaboration that can improve network resource utilization in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a task processing method based on optoelectronic coordination, applied to a controller in a communication system, comprising:

[0006] Receive a task request and determine the target data packet corresponding to the task request, wherein the target data packet carries a task tag;

[0007] In the service chain templates included in the preset service chain template library, feature matching is performed based on the task tag to determine the target service chain template corresponding to the target data packet;

[0008] Based on the service chain identifier of the target service chain template and the task tag, the target data packet is encapsulated to obtain an encapsulated data packet; and based on the encapsulated data packet, an optoelectronic resource association table is obtained.

[0009] The encapsulated data packet is transmitted to the target terminal through the optoelectronic resource association table, optical switch, electrical switch, and transmission path, so that the target terminal can perform task processing based on the encapsulated data packet and obtain task results.

[0010] In one embodiment, the task tag includes one or more of basic metadata, collection communication library metadata, parallel strategy metadata, and requirement metadata; receiving the task request and determining the target data packet corresponding to the task request includes:

[0011] Receive a task request sent by the source terminal and determine the task to be processed corresponding to the task request;

[0012] Based on the task to be processed, the target data packet is obtained; or, based on the multiple data packets corresponding to the task to be processed, the sub-stream is divided to obtain multiple sub-streams, each of which includes multiple target data packets.

[0013] In one embodiment, the service chain template library contains multiple service chain templates, each service chain template including a resource combination and a service strategy; the step of determining the target service chain template corresponding to the target data packet by performing feature matching based on the task tag of the target data packet among the service chain templates contained in the service chain template library includes:

[0014] Metadata fields are extracted from the task tags of the target data packet to obtain tag feature vectors; the matching degree between the tag feature vectors and the tag vectors of each service chain template contained in the service chain template library is calculated to obtain the comprehensive evaluation score of each service chain template.

[0015] The target service chain template is determined based on the comprehensive evaluation score of each service chain template.

[0016] In one embodiment, determining the target service chain template based on the comprehensive evaluation score of each of the service chain templates includes:

[0017] If multiple service chain templates have a comprehensive evaluation score greater than or equal to the first score threshold, the service chain template with the lowest load rate will be selected as the target service chain template; or,

[0018] If all the comprehensive evaluation scores are less than the second score threshold, then the target service chain template is generated based on the task label; if the first score threshold is greater than the second score threshold; or...

[0019] The service chain template with the highest comprehensive evaluation score is selected as the target service chain template.

[0020] In one embodiment, the method further includes:

[0021] Based on the target data packet, multiple candidate paths are determined between the source terminal of the target data packet and the task terminal. The candidate paths include IP segment paths and optical segment candidate paths.

[0022] The path score of each candidate path is obtained by weighting the latency data, energy consumption data, and path cost of each candidate path.

[0023] The transmission path corresponding to the target data packet is determined by filtering based on the path score of each candidate path and preset resource availability conditions.

[0024] In one embodiment, transmitting the encapsulated data packet to the target terminal via the optoelectronic resource association table, optical switch, electrical switch, and transmission path includes:

[0025] Using the source terminal corresponding to the target data packet, the optoelectronic network resources indicated by the optoelectronic resource association table, and the transmission path, each encapsulated data packet is transmitted to the first electrical switch, the optical switch, the second electrical switch, and the target terminal. The target terminal is used to perform task reassembly processing on multiple data packets to obtain a reconstruction task, and to process the reconstruction task to generate a task processing result. The first electrical switch is used to perform data traffic processing and data packet forwarding based on the optoelectronic resource association table.

[0026] In one embodiment, the method further includes:

[0027] If the utilization rate of the optoelectronic network resources indicated by the optoelectronic resource association table exceeds a preset resource threshold, resource reallocation can be performed to obtain reallocated resources, and the reallocated resources are determined to be the optoelectronic network resources indicated by the optoelectronic resource association table; the utilization rate is obtained by real-time detection through resource probes; the optoelectronic network resources include one or more of optical layer, electrical layer port, and electrical layer buffer.

[0028] In one embodiment, the method further includes:

[0029] When the optoelectronic network resources indicated by the optoelectronic resource association table are occupied by lower priority data packets in the same subflow, the optoelectronic network resources occupied by the lower priority data packets are released, and the target data packets are transmitted through the optoelectronic network resources.

[0030] When the target data packet transmission is completed, the lower priority data packets are delayed and retransmitted to ensure that the lower priority data packets are transmitted to the target terminal.

[0031] Secondly, this application also provides a task processing device based on optoelectronic coordination, applied to a controller in a communication system, comprising:

[0032] The first receiving module is used to receive a task request and determine the target data packet corresponding to the task request, wherein the target data packet carries a task tag;

[0033] The matching module is used to perform feature matching based on the task tag among the service chain templates included in the preset service chain template library to determine the target service chain template corresponding to the target data packet;

[0034] The encapsulation module is used to encapsulate the target data packet based on the service chain identifier of the target service chain template and the task tag to obtain an encapsulated data packet; and to obtain an optoelectronic resource association table based on the encapsulated data packet.

[0035] The transmission module is used to transmit the encapsulated data packet to the target terminal through the optoelectronic resource association table, optical switch, electrical switch and transmission path, so that the target terminal can perform task processing based on the encapsulated data packet and obtain task results.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the embodiments of this application.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the embodiments of this application.

[0038] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the embodiments of this application.

[0039] The aforementioned task processing method, apparatus, and computer equipment based on optoelectronic collaboration include: receiving a task request and determining the target data packet corresponding to the task request, wherein the target data packet carries a task tag; performing feature matching based on the task tag among various service chain templates included in a preset service chain template library to determine the target service chain template corresponding to the target data packet; encapsulating the target data packet based on the service chain identifier and task tag of the target service chain template to obtain an encapsulated data packet; obtaining an optoelectronic resource association table based on the encapsulated data packet; and transmitting the encapsulated data packet to the target terminal through the optoelectronic resource association table, an optical switch, an electrical switch, and a transmission path, so that the target terminal can perform task processing based on the encapsulated data packet and obtain the task result. By adopting this method, the task tag of the data packet enables precise binding between the computing power requirement of the data packet and the optoelectronic network resources, thereby improving the utilization rate of optoelectronic network resources while ensuring data transmission performance. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating a task processing method based on optoelectronic coordination in one embodiment;

[0042] Figure 2 This is a flowchart illustrating the steps for obtaining the target data packet in one embodiment;

[0043] Figure 3 This is a flowchart illustrating the steps for determining the target service chain template in one embodiment;

[0044] Figure 4 This is a flowchart illustrating the screening steps in one embodiment;

[0045] Figure 5 This is a flowchart illustrating the transmission steps in one embodiment;

[0046] Figure 6 This is a schematic diagram of the communication architecture in one embodiment;

[0047] Figure 7 This is a flowchart illustrating a task processing method based on optoelectronic coordination in another embodiment;

[0048] Figure 8 This is a flowchart illustrating a task processing method based on optoelectronic coordination in another embodiment;

[0049] Figure 9 This is a structural block diagram of a task processing device based on optoelectronic coordination in one embodiment;

[0050] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0052] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0053] In one embodiment, such as Figure 1 As shown, a task processing method based on optoelectronic coordination is provided. This embodiment applies this method to the controller of a communication system; optionally, the controller can be an SDN controller. This embodiment is illustrated by applying it to an SDN controller. In this embodiment, the task processing method based on optoelectronic coordination includes the following steps:

[0054] Step 102: Receive the task request and determine the target data packet and task tag corresponding to the task request.

[0055] A task request is a request initiated by the source terminal to the SDN controller, such as a large model inference request or a large model training request. The target data packet corresponding to the task request is one or more data packets required to complete the task specified in the request. The task tag carried by the target data packet is globally unique, and there is a one-to-one correspondence between the task tag and the task. The task can be a large model task, which may be a large model inference task or a large model training task. The task tag includes one or more of the following: basic metadata, collection communication library metadata, parallel strategy metadata, and requirement metadata.

[0056] Basic metadata includes task type, task priority, task bandwidth requirements, task latency sensitivity level, task resource preferences, and whether the task crosses a cluster. Collection communication library metadata includes the type and version of the collection communication library. Parallel strategy metadata includes one or more of the following: parallel type, parallelism degree, and communication primitives. Parallel typification includes multiple types such as DP, PP, and TP. Parallelism degree includes 8-GPU DP, 4-stage PP, etc. Communication primitives include All-Reduce, AllGather, etc. Requirements metadata describes the computational power requirements of a task, such as the model and quantity of GPUs required, the memory required, and whether the task crosses a cluster.

[0057] Specifically, the source terminal can send a task request to the controller. The task request may carry a task identifier, etc. Based on the task identifier in the task request, the controller determines the task type and triggers the task tag generation and synchronization process. The controller can determine one or more of the basic metadata, collection communication library metadata, parallel strategy metadata, and requirement metadata corresponding to the task in the task request, and generate a task tag. When the task corresponding to the task request is started, the controller can obtain the task tag corresponding to the task and one or more data packets.

[0058] Step 104: Based on the service chain templates and task tags contained in the preset service chain template library, perform feature matching to determine the target service chain template.

[0059] The service chain template library contains multiple service chain templates, each including resource combinations and service policies. Service chain templates also include dynamic server instantiation, etc. Resource combinations can be combinations of optoelectronic network resources in a communication system, such as optical layer wavelengths (e.g., λ1-λ4), electrical layer ports, and buffer slots (e.g., 200MB buffer for electrical switches and 50MB buffer for optical switches); QoS levels (e.g., packet loss rate ≤0.1% for P0 chains), routing rules (e.g., prioritizing direct optical layer paths), and load balancing methods (e.g., sub-stream fragmentation rules).

[0060] Specifically, the controller can perform structured parsing and feature extraction on the task tags corresponding to the task request to obtain the core metadata fields in the task tags and generate tag feature vectors. The controller can then use the tag vectors corresponding to each service chain template in the service chain template library, calculate the distance between the tag feature vector of the task tag and the tag vectors of each service chain template, and perform matching processing based on the distance to determine the target service chain template corresponding to the task request.

[0061] Step 106: Based on the service chain identifier and task tag of the target service chain template, encapsulate the target data packet to obtain an encapsulated data packet. Then, based on the encapsulated data packet, obtain the optoelectronic resource association table.

[0062] The target data packet can be one or more data packets used to process the large model task corresponding to the task request. The optoelectronic resource association table can be a resource binding table, that is, each service chain specifies or binds optical network resources and / or electrical network resources respectively.

[0063] Specifically, after determining the target service chain template corresponding to the task request, the controller can concatenate the identifier of the target service chain template with the task tag to obtain a concatenation field. This concatenation field is then used to encapsulate the header of each target data packet, resulting in multiple encapsulated data packets corresponding to the task request. Based on the identifier of the target service chain template and the task tag corresponding to the task request, the header of the target data packet is encapsulated to obtain the encapsulated target data packet. This allows for the generation of a transmission queue and an optoelectronic resource association table based on the multiple encapsulated data packets. The optoelectronic resource association table represents the optical network resources and / or electrical network resources bound to each data packet.

[0064] Step 108: Transmit encapsulated data packets to the target terminal through the optoelectronic resource association table, optical switch, electrical switch, and transmission path, so that the target terminal can perform task processing based on the encapsulated data packets and obtain task results.

[0065] Specifically, the controller can send the generated optoelectronic resource association table to various network devices, such as source terminals, optical switches, and electrical switches, so that each network device can transmit encapsulated data packets based on the optoelectronic resources indicated in the table. For example, when a source terminal initiates a data transmission request, the data packet is transmitted from the source terminal to the electrical switch, and then from the electrical switch to the optical switch, using the optoelectronic resources indicated in the table. The encapsulated data packet is then transmitted from the optical switch to the electrical switch in the target data center, and finally to the target terminal, using the same optoelectronic resources. The target terminal can then perform task processing based on receiving at least one encapsulated data packet, such as large model training or large model inference, to obtain the task result.

[0066] In the aforementioned task processing method based on optoelectronic collaboration, a task request is received, and the target data packet corresponding to the task request is determined. The target data packet carries a task tag. Among the service chain templates included in the preset service chain template library, feature matching is performed based on the task tag to determine the target service chain template corresponding to the target data packet. Based on the service chain identifier and task tag of the target service chain template, the target data packet is encapsulated to obtain an encapsulated data packet. Based on the encapsulated data packet, an optoelectronic resource association table is obtained. Through the optoelectronic resource association table, optical switches, electrical switches, and transmission paths, the encapsulated data packet is transmitted to the target terminal, enabling the target terminal to perform task processing based on the encapsulated data packet and obtain the task result. By adopting this method, the task tag of the data packet enables precise binding between the computing power requirements of the data packet and optoelectronic network resources, improving the utilization rate of optoelectronic network resources while ensuring data transmission performance.

[0067] In one embodiment, the task tag includes one or more of the following: basic metadata, collection communication library metadata, parallel strategy metadata, and requirement metadata. For example... Figure 2 As shown, the specific processing steps of "receiving a task request and determining the target data packet corresponding to the task request" include:

[0068] Step 202: Receive the task request sent by the source terminal and determine the task to be processed corresponding to the task request.

[0069] The task to be processed can be the task requested to be executed in the task request, such as a task associated with a large model, such as a large model inference task or a large model training task.

[0070] Specifically, the source terminal can send a task request to the controller, which may include a task identifier, etc. The controller determines the task type and the task to be processed based on the task identifier in the task request.

[0071] Step 204: Obtain the target data packet based on the task to be processed. Alternatively, divide the multiple data packets corresponding to the task to be processed into sub-streams to obtain multiple sub-streams, each of which includes multiple target data packets.

[0072] Specifically, when the controller determines that a pending task corresponding to a task request has started, it can acquire one or more data packets corresponding to that task, i.e., acquire one or more data packets required to execute the task. If only one data packet exists, it can be identified as the target data packet. If multiple data packets are involved in the task, the controller can divide the multiple data packets into multiple sub-streams according to preset sub-stream fragmentation rules. Each sub-stream can include multiple data packets, and the sub-streams can be transmitted or arranged in a preset order or generation order. Each sub-stream is bound to independent optoelectronic resources, which avoids single-path congestion.

[0073] In this embodiment, the rational allocation of optoelectronic resources can be achieved in scenarios involving the transmission of multiple data packets, enabling the optoelectronic device to accurately identify the resource requirements of the tasks to be processed by the data packets.

[0074] In one embodiment, the service chain template library contains multiple service chain templates, each service chain template including resource combinations and service strategies. The specific processing steps for "determining the target service chain template by performing feature matching based on the service chain templates and task tags contained in the preset service chain template library" include:

[0075] Metadata fields are extracted from the task tags of the target data packet to obtain tag feature vectors. The matching degree between each tag feature vector and the tag vector of each service chain template in the service chain template library is calculated to obtain a comprehensive evaluation score for each service chain template. Based on the comprehensive evaluation scores of each service chain template, the target service chain template is determined.

[0076] The service chain template library contains multiple service chain templates, each including resource combinations and service policies. Service chain templates also include dynamic server instantiation, etc. Resource combinations can be combinations of optoelectronic network resources in a communication system, such as optical layer wavelengths (e.g., λ1-λ4), electrical layer ports, and buffer slots (e.g., 200MB buffer for electrical switches and 50MB buffer for optical switches); QoS levels (e.g., packet loss rate ≤0.1% for P0 chains), routing rules (e.g., prioritizing direct optical layer paths), and load balancing methods (e.g., sub-stream fragmentation rules).

[0077] Specifically, the controller can perform structured parsing and feature extraction on the task tags corresponding to the task requests to obtain the core metadata fields in the task tags and generate tag feature vectors. The controller can then use the tag vectors corresponding to each service chain template in the service chain template library, and calculate the distances between the tag feature vector of the task tag and the tag vectors of each service chain template. Based on these distances and pre-set first and second score thresholds, the controller filters to obtain the target service chain template corresponding to the task tag. The distances can be determined based on preset cosine distance algorithms or sine distance algorithms, and the controller can use these distances as a comprehensive evaluation score. The target service chain template can be from a preset service chain template library or can be generated in real-time.

[0078] The task tags can be shown in Table 1 below:

[0079]

[0080] In this embodiment, by performing structured parsing and field matching on the task tags, a resource service chain template that accurately matches each data packet corresponding to the task tag can be obtained, providing a data foundation for the subsequent rational allocation of resources and further improving resource utilization.

[0081] In one embodiment, such as Figure 3 As shown, the specific processing steps for the step "determine the target service chain template based on the comprehensive evaluation scores of each service chain template" include:

[0082] Step 302: If multiple service chain templates have a comprehensive evaluation score greater than or equal to the first score threshold, the service chain template with the lowest load rate is selected as the target service chain template. Alternatively,

[0083] Specifically, the controller determines the distance between each service chain template in the preset service chain template library and the task tag, and determines this distance as the comprehensive evaluation score between the service chain template and the task tag. The controller can compare the comprehensive evaluation score corresponding to each service chain template with a preset first score threshold. If it is determined that there are multiple service chain templates with a comprehensive evaluation score greater than or equal to the first score threshold, the controller can further determine the load rate of the service chain templates with a load rate greater than or equal to the first score threshold, and determine the service chain template with the lowest load rate as the target service chain template.

[0084] Step 304: If all comprehensive evaluation scores are less than the second score threshold, then generate a target service chain template based on the task label. The first score threshold is greater than the second score threshold. Alternatively,

[0085] Specifically, the controller can compare the comprehensive evaluation score corresponding to each service chain template with a preset second score threshold. If it is determined that the current comprehensive evaluation scores of each service chain template are all less than the preset second score threshold, the controller can trigger the service chain template generation mechanism. Through this generation mechanism, the resource combination and service strategy corresponding to the service chain template are adjusted based on the task tag to obtain the adjusted service chain template, and the adjusted service chain template is determined to be the target service chain template that matches the task tag.

[0086] Step 306: Determine the service chain template with the highest comprehensive evaluation score as the target service chain template.

[0087] Specifically, if the controller determines that there are no multiple comprehensive evaluation scores between each service chain template and the task tag in the preset service chain template library, and there are no service chain templates with comprehensive evaluation scores greater than or equal to the first score threshold, and there are no service chain templates with comprehensive evaluation scores less than the second score threshold, the controller can select the service chain template with the highest comprehensive evaluation score as the target service chain template.

[0088] In this embodiment, by using the comprehensive evaluation score based on the service chain template for precise matching, the adaptability of the service chain template bound to the task tag can be further improved, providing a foundation for the subsequent collaborative use of optoelectronic resources, and also realizing the precise association between task tags, service chain templates and optoelectronic resources.

[0089] In one embodiment, such as Figure 4 As shown, this optoelectronic collaborative task processing method also includes:

[0090] Step 402: Based on the target data packet, determine multiple candidate paths between the source terminal and the task terminal of the target data packet. The candidate paths include IP segment paths and optical segment candidate paths.

[0091] Specifically, the controller can obtain the IP layer topology of the communication system through BGP-LS and the optical layer topology through GMPLS. The IP layer topology and the optical layer topology are then fused to obtain the communication topology corresponding to the communication system. The IP layer topology may include AS domains and routing prefixes, the optical layer topology may include fiber optic links and wavelength resources, and the communication topology may include information such as routers, switches, IP routes, and optical wavelengths. Based on the sender and receiver corresponding to the target data packet, the controller can extract multiple candidate paths corresponding to the target data packet from this communication topology. These candidate paths include IP segment paths in the IP layer topology and optical segment candidate paths in the optical layer topology.

[0092] Step 404: The candidate paths are weighted based on their latency data, energy consumption data, and path cost to obtain their path scores.

[0093] Among them, the latency data can be the end-to-end total latency, which can be determined by the sum of optical path latency, IP path latency, and photoelectric conversion latency; the energy consumption data represents the end-to-end total energy consumption, including optical path energy consumption and IP path energy consumption; the path cost includes optical path cost and IP path cost.

[0094] Specifically, for each candidate path, the controller can obtain the sum of the optical path delay, IP path delay, and photoelectric conversion delay corresponding to that candidate path. The optical path delay includes fiber optic transmission delay; the IP path delay includes IP port processing delay and IP link transmission delay; the IP port processing delay can be as long as a router forwards a data packet (e.g., 1 μs); the photoelectric conversion delay is a fixed value, added once for each photoelectric connection node. This embodiment does not limit the above fixed value and can be determined based on the actual application scenario. The controller can determine the weighting factors for delay data, energy consumption data, and path cost based on the actual application scenario, and perform weighted processing based on the delay data, energy consumption data, path cost, and the corresponding weighting factors to obtain the path score corresponding to the candidate path.

[0095] Step 406: Based on the path score of each candidate path and the preset resource availability conditions, filter to determine the transmission path corresponding to the target data packet.

[0096] The preset resource availability condition can be that all types of resources are available.

[0097] Specifically, the controller can filter candidates based on their path scores from highest to lowest, and determine the candidate path with the highest score and all types of resources being available as the transmission path corresponding to the target data packet.

[0098] In this embodiment, the dynamic orchestration of optoelectronic resources can be achieved by combining the priority of the service chain, thereby further improving the utilization rate of path resources.

[0099] In one embodiment, the specific processing steps of "transmitting encapsulated data packets to the target terminal through the optoelectronic resource association table, optical switch, electrical switch, and transmission path" include:

[0100] Using the source terminal corresponding to the target data packet, the optoelectronic network resources indicated by the optoelectronic resource association table, and the transmission path, each encapsulated data packet is transmitted to the first electrical switch, the optical switch, the second electrical switch, and the target terminal. The target terminal is used to perform task reassembly processing on multiple data packets to obtain a reconstruction task, and then processes the reconstruction task to generate a task processing result. The first electrical switch is used for data traffic processing and data packet forwarding based on the optoelectronic resource association table.

[0101] Among them, the first electrical switch can be an access layer electrical switch; the second electrical switch can be an electrical switch in the target data center; and the optical switch can be a switch in the optical layer / optical topology.

[0102] Specifically, the data transmission request initiated by the source terminal can transmit the encapsulated data packet to the access layer electrical switch after the first data packet is sent from the source terminal. The electrical switch performs local traffic aggregation, traffic shaping, and preliminary forwarding based on the tag information in the data packet header to ensure that the data packet meets the stability requirements of optical layer transmission. In this way, the encapsulated data packet can be transmitted to the optical switch in cross-cluster / cross-DC scenarios. The optical switch performs direct optical layer transmission, buffer management, and cross-optical layer transmission of the encapsulated data packet based on the optical network resources indicated by the optoelectronic resource association table. Then, the encapsulated data packet can be transmitted to the electrical switch of the target data center and forwarded to the target terminal. The target terminal needs to reassemble the multiple sub-streams and multiple data packets into complete task data based on the tag information to ensure that the large model task can be processed normally.

[0103] In this embodiment, the sending end selects the bound optoelectronic resource for transmission according to the binding table, and the receiving end routes the sub-stream to the same virtual channel according to the task tag, thereby achieving accurate transmission of data packets.

[0104] In one embodiment, the optoelectronic-coordinated task processing method further includes:

[0105] When the utilization rate of optoelectronic network resources indicated by the optoelectronic resource association table exceeds a preset resource threshold, resource reallocation can be performed to obtain reallocated resources, which are then identified as optoelectronic network resources indicated by the optoelectronic resource association table. The utilization rate is obtained in real time through resource probes. Optoelectronic network resources include one or more of the following: optical layer, electrical layer ports, and electrical layer buffers.

[0106] Specifically, when transmitting data packets based on the utilization rate of optoelectronic network resources indicated by the optoelectronic resource association table, if it is determined that the current resource utilization rate has exceeded a preset threshold, the controller can perform resource reallocation, identifying other available resources of the same type as the reallocated resources. In this way, optical switches and electrical switches can continue to transmit other encapsulated data packets through these reallocated resources, ensuring that the transmission delay of each encapsulated data packet corresponding to the task to the target terminal is less than a preset delay threshold, such as 10µs.

[0107] In this embodiment, the timely reallocation of optoelectronic resources used by the service chain can realize the migration of optoelectronic resources used by data packets of some sub-streams from high-load resources to low-load resources, reduce the transmission latency of the same batch of data packets corresponding to the task, ensure the synchronization of the transmission of each data packet, ensure the timely processing of large model tasks, and also ensure the accuracy of large model task processing.

[0108] In one embodiment, such as Figure 5 As shown, this optoelectronic collaborative task processing method also includes:

[0109] Step 502: When the optoelectronic network resources indicated by the optoelectronic resource association table are occupied by lower priority data packets in the same subflow, release the optoelectronic network resources occupied by the lower priority data packets and transmit the target data packets through the optoelectronic network resources.

[0110] Among them, lower priority data packets can be data packets with a lower priority than the currently transmitted encapsulated data packets. For example, the priority of data packets in the control chain sub-flow in the P0 service chain is higher than the priority of the inference flow in the P1 service chain.

[0111] Specifically, the SDN controller monitors resource usage in real time through sensors on optical / electrical switches. For example, it can be determined that the optical layer λ1 of the P1 chain is 80% occupied, while the control chain requires λ1. When a conflict is detected, "service chain preemption" is triggered, suspending the transmission of low-priority sub-streams in the P1 chain, such as sub-streams of the P1 inference stream, releasing the optoelectronic resources they occupy (such as the 20Gbps bandwidth of electrical layer port B), and migrating the control chain to the released resources.

[0112] Step 504: When the target data packet transmission is completed, the lower priority data packets are delayed and retransmitted to ensure that the lower priority data packets are transmitted to the target terminal.

[0113] Specifically, after the control chain transmission is completed, the preempted resources are released, and the suspended P1 chain sub-streams notify the receiver to delay processing (e.g., 10ms delay) through the ROCEv2 DCQCN mechanism, and the data during the suspension period is completed through the retransmission mechanism (only critical sub-streams are retransmitted); to ensure that high-priority communication algorithms are processed first, the ECN of the queue where the high-priority communication algorithm is located can be disabled as needed; in addition, during retransmission, the originally bound optoelectronic resources are used first (if available), otherwise the second-best resources in the same service chain (e.g., spare optical wavelengths) are selected.

[0114] In this embodiment, by adjusting resources in a timely manner between different resource chains, the real-time transmission and real-time performance of high-priority data packets can be guaranteed, and the timely transmission of data packets that are paused can be achieved through a data retransmission mechanism.

[0115] The following describes in detail the specific implementation steps of the above-mentioned task processing method based on optoelectronic coordination, with reference to a specific embodiment:

[0116] To meet the requirements of high latency, high bandwidth, and high reliability of distributed intelligent computing networks due to the high burstiness and strong synchronization of large-scale business traffic, this embodiment focuses on a collaborative bearer network of electrical and optical switches, proposing a wide-area intelligent computing network method and system based on task tag-based optoelectronic switching collaborative optimization. In the optoelectronic collaborative network, the electrical switch is responsible for local traffic aggregation and forwarding cross-rack traffic to the optical switch; the optical switch is responsible for large-scale traffic scheduling. After receiving cross-rack traffic forwarded by the electrical switch, it realizes long-distance transmission between server rooms and data centers through direct optical layer connection or wide-area optical equipment, utilizing the low latency characteristics of the optical layer to reduce end-to-end latency. To better achieve optoelectronic collaboration, the specific technical solutions in this embodiment include: task tag-driven resource service chain construction, service chain-based optoelectronic resource collaborative scheduling, and optoelectronic collaborative caching based on communication and computing collaboration, such as... Figure 6 As shown, this includes a DC-OCS plane and multiple POD networks. Each POD network includes multiple interconnected leaf switches, spine switches, and multiple servers.

[0117] The process of building a task-label-driven resource service chain includes: assigning globally unique task labels (such as the hash value of TASK_LLAMA7B_TRAIN) to large model tasks. These labels contain basic metadata such as task type (training / inference, etc.), priority (P0 / P1 / P2), bandwidth requirement (100Gbps), latency sensitivity (high / medium / low), resource preference (such as "priority optical layer direct connection wavelength λ1"), and cross-cluster markers; collecting communication library metadata such as the type of communication library (such as NCCL / Megatron-LM / DeepSpeed) and version; and parallelism type (DP / PP). Metadata for parallel strategies such as parallelism (e.g., 8-card DP, 4-stage PP) and communication primitives (All-Reduce / AllGather); metadata describing computing power requirements (e.g., GPU model, quantity, memory, whether it crosses clusters); and a pre-defined service chain template library: This library supports multiple parallel strategies such as data parallelism (DP), pipelined parallelism (PP), and tensor parallelism (TP), performs algorithm tagging for the aggregated communication library, transforms large-scale parallel algorithms into point-to-point (P2P) communication traffic, and designs network service chains to match the different stages of the intelligent computing service lifecycle. The SDN controller predefines multiple types of resource service chain templates. Each template includes resource combinations, service strategies, and dynamic service chain instantiation. Resource combinations include optical layer wavelengths (e.g., λ1-λ4), electrical layer ports, cache slots (e.g., 200MB cache for electrical switches, 50MB cache for optical switches); QoS levels (e.g., packet loss rate ≤0.1% for P0 chains), routing rules (e.g., prioritizing direct optical layer paths), and load balancing methods (e.g., sub-stream fragmentation rules).

[0118] When a task starts, the controller matches the closest service chain template based on the task tag (e.g., TASK_LLAMA_TRAIN matches “P0 training chain”) and dynamically instantiates it into a specific resource service chain (e.g., the instantiated service chain contains optical layer λ1 + electrical layer port A + 200MB electrical switch cache); when the task ends or is scaled down, the corresponding service chain is destroyed and the resources are released.

[0119] Specifically, the service chain template matching process in the template library includes: Upon task startup, the SDN server controller first performs structured parsing of the task labels, extracts core metadata fields related to service chain template adaptation, and parses the output label feature vector. For example, it could be training / P0, 100Gbps / high latency sensitive, NCCL2.18, TP-8 card, priority λ1-λ4, cross-DC. Then, it constructs a service chain module library for the multi-parallel strategies involved in the task. The specific modeling rules are as follows: "TP parallel P0..." Taking the training template (template ID: LLAM-Train-TP-P0) as an example, the feature vector can be: adaptation task = training / P0 (resource exclusive), resource combination = λ1 + λ2 (50Gbps per wavelength) + electrical port A (20Gbps spare), QoS = packet loss rate ≤ 0.1% / latency ≤ 15ms, adaptation communication library = NCCL2.10+, parallel adaptation = TP-4~16 cards, load balancing = sub-stream fragmentation (4 sub-streams), cross-DC support = yes], and at the same time, the current resource availability status of the template is synchronized in real time (such as whether the template-bound λ1 is idle, whether the electrical port bandwidth is sufficient; this can provide a basis for judging resource availability for subsequent matching). Secondly, the matching metric calculation of tags and module library is performed, which can be divided into exact matching, interval matching, partial matching, and no matching according to the degree of matching. The controller sorts the comprehensive matching degree from high to low, combined with the template resources Based on the availability of the source, the target service chain template is ultimately determined. For example, a template with a "comprehensive matching score ≥ 80" and "core resources (optical wavelength / electrical port) completely idle" is prioritized. If multiple templates with a "comprehensive matching score ≥ 80" exist, the current resource load of the module is compared, and the service chain template with the lower load rate is prioritized. If all templates have a "comprehensive matching score < 60" (no suitable template), a "dynamic template generation mechanism" is triggered to adjust their resource combinations and strategies, generating a "temporary dedicated template" to ensure task adaptation. Finally, after selecting the closest service chain template, the controller synchronizes the "template ID - resource binding rules" to the task source terminal (such as the GPU server) and network devices (electrical / optical switches), and instantiates a specific resource service chain based on the template (such as binding λ1 + electrical port A + 200MB electrical cache), completing the closed-loop association of "tag - template - resource". Specifically, for example... Figure 7 As shown:

[0120] 1. The terminal submits a task request to the SDN controller, triggering task tag generation and synchronization.

[0121] 2. Then, the SDN controller allocates a dedicated resource service chain to the task through "label-template matching", which clarifies the combination of optoelectronic resources and forwarding strategy required for data packet transmission. In the case of multi-data packet transmission scenario (such as 100 training data packets), the controller divides the data packets into N sub-streams (such as 4 sub-streams, each with 25 data packets) according to the "sub-stream fragmentation rule". Each sub-stream is bound to independent optoelectronic resources to avoid single-path congestion.

[0122] 3. After the large model task starts, it enters the data processing stage, generating multiple data packets to be transmitted. The source terminal needs to encapsulate the "task tag + service chain identifier" into the data packet header, enabling network devices to identify the task requirements and resource binding relationships of the data packets in real time. After encapsulation, the data packets enter the "transmission queue," and the source terminal initiates transmission requests according to the priority in the "resource binding table."

[0123] 4. After the data packet is sent from the source terminal, it first enters the access layer electrical switch. Based on the tag information in the data packet header, the electrical switch performs local traffic aggregation, traffic shaping and preliminary forwarding to ensure that the data packet meets the "stability requirements" of optical layer transmission.

[0124] 5. After data packets from across clusters / DCs enter the optical switch, the optical switch performs direct optical layer transmission and buffer management based on the "resource binding table" of the tag information.

[0125] 6. Data packets transmitted across the optical layer eventually reach the electrical switch in the target data center and are then forwarded to the target terminal. The target terminal needs to reassemble the multiple sub-streams and multiple data packets into complete task data based on the tag information to ensure that the large model task can be processed normally.

[0126] The process of the service chain-based optoelectronic resource collaborative scheduling mechanism includes:

[0127] Based on the perceived status of the optical switching network and Ethernet network, and combined with the priority of service chains, the SDN controller dynamically optimizes resource orchestration and scheduling. Therefore, the centralized control architecture of the wide-area intelligent computing network (SDN) includes: optoelectronic topology discovery and resource awareness, path calculation and conflict-free routing address adjustment, and dynamic adjustment.

[0128] The joint topology discovery can be achieved by the SDN controller obtaining the IP layer topology (such as AS domains and routing prefixes) through BGP-LS and the optical layer topology (such as fiber optic links and wavelength resources) through GMPLS, and then merging them to generate an "IP-optical layer joint topology map" (containing information such as routers, switches, IP routes, and optical wavelengths).

[0129] Joint path calculation can be based on a joint topology graph, combined with multiple constraints of latency, energy consumption, and cost when selecting a path, to calculate the shortest end-to-end path. Based on priority, it can determine whether to choose the path that is "ideal for IP logic (e.g., the shortest AS path) and has available optical layer resources (e.g., wavelength λ1 is idle)" or to degrade to the path that is suboptimal for IP but has available optical layer resources.

[0130] Furthermore, a unique identifier is assigned to the selected joint path, and an "Optical Tag" is extended to the IP packet header to guide the optical switch to forward the packet along that path (e.g., selecting wavelength λ1 via WSS). In the event of path failure, IP layer rerouting and optical layer path rebinding are automatically triggered. The total end-to-end delay can be determined by the sum of the optical path delay, IP path delay, and optoelectronic conversion delay.

[0131] Optical path latency includes fiber optic transmission latency; IP path latency includes IP port processing latency (e.g., 1μs for a router to forward a data packet) and IP link transmission latency; optoelectronic conversion latency is a fixed value, added once for each optoelectronic connection node. End-to-end total energy consumption = optical path energy consumption + IP path energy consumption, with lower optical path energy consumption and higher IP layer energy consumption; end-to-end total cost = optical path cost + IP path cost, with optical resource occupation cost being higher than IP port occupation cost; thus, based on the priority of task tags, the priorities of latency, energy consumption, and cost are dynamically adjusted to ensure that the core requirements of high-priority tasks are met first.

[0132] Then, the multi-objective cooperative path calculation process is performed, as follows:

[0133] S1, Generate candidate path set: Based on the joint topology graph of IP and optical, the SDN controller generates IP segment and optical segment candidate paths for the "source node-target node" of the task.

[0134] S2, Multi-Constraint Evaluation: For each candidate path, calculate the comprehensive score according to the "constraint priority weight".

[0135] S3, Optimal Path Selection: Prioritize the path with the highest overall score and all available resources. If the optimal path's resources are unavailable, select the second-best path based on the overall score.

[0136] To better serve intelligent computing network control, the service chain-based optoelectronic resource collaborative scheduling mechanism includes steps such as service chain priority allocation, cross-layer resource binding, service chain load awareness, and service chain load balancing. Figure 8 As shown, the details are as follows:

[0137] Service chain priority allocation, i.e. service chain resolution: Based on the priority of task tags (P0 / P1), global priority is assigned to resource service chains (e.g., P0 chain > P1 chain). For example, the control plane data association has the highest priority "control chain" and exclusively occupies dedicated resources; ALLtoALL communication has a higher priority than ALLReduce, and TP parallel communication has a higher priority than PP parallel and DP parallel methods.

[0138] Cross-layer resource binding, or resource allocation strategy, involves splitting a single large synchronous flow (e.g., 100Gbps) into N sub-flows (e.g., 4 sub-flows, each 25Gbps). Each sub-flow is assigned a unique sequence number (SN) and path identifier (PathID). Each resource service chain instance contains an "optical-electrical resource binding table," clearly defining the correspondence between sub-flows and optical-electrical resources (e.g., sub-flow 1 → optical layer λ1 + electrical layer port A, sub-flow 2 → optical layer λ2 + electrical layer port B), avoiding congestion on a single path. The SDN controller monitors the utilization of each link in real time and directs new traffic to the path with the lowest utilization. The sending end selects the bound optical-electrical resources for transmission according to the binding table, and the receiving end routes the sub-flows to the same virtual channel based on the task tag.

[0139] Service chain load awareness: The SDN controller deploys lightweight probes (such as OAM packets embedded in optical switches and sFlow of electrical switches) within the service chain to monitor the utilization of each resource in real time (such as 70% utilization of optical layer λ1 and 80% load of electrical layer port A) and latency (such as 5ms latency of optical layer λ1 and 2ms latency from electrical layer port A to optical switch).

[0140] Intra-service load balancing: When the utilization rate of a certain resource exceeds the threshold (e.g., optical layer λ1 reaches 80%), the controller triggers intra-service resource reallocation: some sub-streams are migrated from high-load resources to low-load resources in the same service chain (e.g., from optical layer λ1 to electrical layer port C), ensuring that the latency difference of all sub-streams in the service chain is ≤10μs (meeting the synchronization accuracy requirements of large models).

[0141] Collision Detection and Preemption: The SDN controller monitors resource usage in real time through sensors on optical / electrical switches (e.g., optical layer λ1 of P1 chain is 80% occupied, while the control chain requires λ1). When a collision is detected, "service chain preemption" is triggered, suspending the transmission of low-priority sub-streams in P1 chain (e.g., sub-streams of P1 inference streams), releasing the optoelectronic resources they occupy (e.g., the 20Gbps bandwidth of electrical layer port B), and migrating the control chain to the released resources to ensure real-time performance (latency ≤ 1ms). After the control chain transmission is completed, the preempted resources are released, and the suspended P1 chain sub-streams notify the receiver of delayed processing (e.g., delay 10ms) through the ROCEv2 DCQCN mechanism, and complete the data during the suspension period through the retransmission mechanism (only critical sub-streams are retransmitted). To ensure that high-priority communication algorithms are processed first, the ECN of the queue where the high-priority communication algorithm is located can be disabled as needed. In addition, during retransmission, the originally bound optoelectronic resources are used first (if available), otherwise the second-best resource within the same service chain (e.g., a spare optical wavelength) is selected.

[0142] Dynamic Tag Update Mechanism: The SDN controller integrates a lightweight probe to collect key network status indicators in real time, including bandwidth utilization, latency, and resource load. Bandwidth utilization includes current bandwidth demand and actual bandwidth usage; latency includes optical layer transmission latency and electrical layer processing latency; and resource load includes optical wavelength usage and switch port congestion. Corresponding thresholds are set for different indicators, and the system automatically determines whether to update the tag based on these preset thresholds. Once an indicator exceeds a threshold, the corresponding part of the tag is triggered for dynamic modification, and the changes are synchronized to network devices in real time through extended IP header fields. The globally unique hash value remains unchanged during the update process to ensure task identifier consistency. The SDN controller then re-matches the service chain template library based on the updated tag.

[0143] Because optical switches have limited caching capacity, this embodiment will address the issue of buffer overflow caused by sudden surges in large model traffic (such as surges during training task synchronization) through communication and computing collaboration, as well as optoelectronic collaboration. The process of the optoelectronic collaborative caching scheme based on communication and computing collaboration includes:

[0144] A traffic shaper is deployed at the egress direction of the electrical switch to smooth burst traffic (e.g., 100Gbps bursts) to a fixed rate (e.g., 80Gbps), preventing instantaneous overload of the optical switch. Simultaneously, computational latency is used to mask communication latency, avoiding increased communication latency overhead after traffic shaping. The optical switch buffer is divided into static and dynamic zones. The static zone is bound to fixed resources and reserved for high-priority tasks; resources are requested through predictive bandwidth reservation before task initiation. The dynamic zone uses the WRED (Weighted Random Early Detection) method; when the load or resource utilization exceeds a threshold, low-priority traffic is discarded in advance, preserving buffer space for high-priority traffic (e.g., training traffic). Collaborative caching with the electrical switch is performed when necessary. The sizes of the static and dynamic zones can be adjusted according to task requirements.

[0145] Since the traffic during large model training is predictable, dedicated cache slots are reserved in advance on the optical switch based on traffic characteristics. Before the task starts, fixed bandwidth and cache resources are requested from the optical switch to ensure that critical traffic is used first. When the optical switch detects that the optical switch's cache is tight, it temporarily stores some traffic to be forwarded and waits for the optical switch to finish processing before forwarding it, thus relieving the pressure on the optical switch.

[0146] The method provided in this embodiment can drive the dynamic construction and optoelectronic collaborative scheduling of resource service chains through task tags (including multi-dimensional metadata such as task type, priority, bandwidth requirements, latency sensitivity, parallel strategy, aggregated communication library, and computing power requirements); it dynamically instantiates dedicated resource combinations (optical wavelength + electrical port + cache) based on the service chain template library to achieve precise binding of computing power and network resources; it adopts sub-stream sharding and cross-layer resource mapping mechanisms, combined with real-time load perception (probe monitoring utilization / latency) to trigger load balancing within the service chain (migrating sub-streams when resource utilization exceeds the threshold, ensuring latency difference ≤10μs); it resolves conflicts through priority preemption mechanisms (such as pausing low-priority sub-streams to release resources) and optoelectronic collaborative caching schemes (static area reserving high-priority traffic + dynamic area WRED discarding strategy); and it integrates a dynamic tag update mechanism (automatically upgrading and reconfiguring service chains when bandwidth requirements exceed the threshold) to ensure that resource scheduling always matches task requirements, ultimately achieving efficient, low-latency, and highly reliable optoelectronic collaborative carrying of wide-area intelligent computing networks.

[0147] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0148] Based on the same inventive concept, this application also provides an optoelectronic collaborative task processing device for implementing the optoelectronic collaborative task processing method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more optoelectronic collaborative task processing device embodiments provided below can be found in the limitations of the optoelectronic collaborative task processing method described above, and will not be repeated here.

[0149] In one exemplary embodiment, such as Figure 9 As shown, a task processing device 900 based on optoelectronic coordination is provided, which is applied to the controller of a communication system and includes:

[0150] The first receiving module 902 is used to receive a task request and determine the target data packet corresponding to the task request, wherein the target data packet carries a task tag;

[0151] The matching module 904 is used to perform feature matching based on the task tag among the service chain templates included in the service chain template library to determine the target service chain template corresponding to the target data packet;

[0152] The encapsulation module 906 is used to encapsulate the target data packet based on the service chain identifier of the target service chain template and the task tag to obtain an encapsulated data packet; and to obtain an optoelectronic resource association table based on the encapsulated data packet.

[0153] The transmission module 908 is used to transmit the encapsulated data packet to the target terminal through the optoelectronic resource association table, optical switch, electrical switch and transmission path, so that the target terminal can perform task processing based on the encapsulated data packet and obtain task results.

[0154] Each module in the aforementioned optoelectronic collaborative task processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0155] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores task data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements a task processing method based on optoelectronic coordination.

[0156] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0157] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0158] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0159] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0160] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0161] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0162] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0163] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A task processing method based on optoelectronic coordination, characterized in that, The method, applied to a controller in a communication system, includes: Receive a task request and determine the target data packet corresponding to the task request, wherein the target data packet carries a task tag; The target service chain template is determined by feature matching based on the service chain templates contained in the preset service chain template library and the task tags. Based on the service chain identifier of the target service chain template and the task tag, the target data packet is encapsulated to obtain an encapsulated data packet; and based on the encapsulated data packet, an optoelectronic resource association table is obtained. The encapsulated data packet is transmitted to the target terminal through the optoelectronic resource association table, optical switch, electrical switch, and transmission path, so that the target terminal can perform task processing based on the encapsulated data packet and obtain task results.

2. The method according to claim 1, characterized in that, The task tags include one or more of the following: basic metadata, collection communication library metadata, parallel strategy metadata, and requirement metadata; The step of receiving a task request and determining the target data packet corresponding to the task request includes: Receive a task request sent by the source terminal and determine the task to be processed corresponding to the task request; Based on the task to be processed, the target data packet is obtained; or, based on the multiple data packets corresponding to the task to be processed, the sub-stream is divided to obtain multiple sub-streams, each of which includes multiple target data packets.

3. The method according to claim 2, characterized in that, The service chain template library contains multiple service chain templates, each of which includes a resource combination and a service strategy; the step of determining the target service chain template by performing feature matching based on the service chain templates in the preset service chain template library and the task tags includes: Metadata fields are extracted from the task tags to obtain tag feature vectors; the matching degree between the tag feature vectors and the tag vectors of each service chain template contained in the service chain template library is calculated to obtain the comprehensive evaluation score of each service chain template. The target service chain template is determined based on the comprehensive evaluation score of each service chain template.

4. The method according to claim 1, characterized in that, The determination of the target service chain template based on the comprehensive evaluation score of each service chain template includes: If multiple service chain templates have a comprehensive evaluation score greater than or equal to the first score threshold, the service chain template with the lowest load rate will be selected as the target service chain template; or, If all the comprehensive evaluation scores are less than the second score threshold, then the target service chain template is generated based on the task label; if the first score threshold is greater than the second score threshold; or... The service chain template with the highest comprehensive evaluation score is selected as the target service chain template.

5. The method according to claim 1, characterized in that, The method further includes: Based on the target data packet, multiple candidate paths are determined between the source terminal of the target data packet and the task terminal. The candidate paths include IP segment paths and optical segment candidate paths. The path score of each candidate path is obtained by weighting the latency data, energy consumption data, and path cost of each candidate path. The transmission path corresponding to the target data packet is determined by filtering based on the path score of each candidate path and preset resource availability conditions.

6. The method according to claim 1, characterized in that, The process of transmitting the encapsulated data packet to the target terminal through the optoelectronic resource association table, optical switch, electrical switch, and transmission path includes: Using the source terminal corresponding to the target data packet, the optoelectronic network resources indicated by the optoelectronic resource association table, and the transmission path, each encapsulated data packet is transmitted to the first electrical switch, the optical switch, the second electrical switch, and the target terminal. The target terminal is used to perform task reassembly processing on multiple data packets to obtain a reconstruction task, and to process the reconstruction task to generate a task processing result. The first electrical switch is used to perform data traffic processing and data packet forwarding based on the optoelectronic resource association table.

7. The method according to claim 6, characterized in that, The method further includes: If the utilization rate of the optoelectronic network resources indicated by the optoelectronic resource association table exceeds a preset resource threshold, resource reallocation can be performed to obtain reallocated resources, and the reallocated resources are determined to be the optoelectronic network resources indicated by the optoelectronic resource association table; the utilization rate is obtained by real-time detection through resource probes; the optoelectronic network resources include one or more of optical layer, electrical layer port, and electrical layer buffer.

8. The method according to claim 6, characterized in that, The method further includes: When the optoelectronic network resources indicated by the optoelectronic resource association table are occupied by lower priority data packets in the same subflow, the optoelectronic network resources occupied by the lower priority data packets are released, and the target data packets are transmitted through the optoelectronic network resources. When the target data packet transmission is completed, the lower priority data packets are delayed and retransmitted to ensure that the lower priority data packets are transmitted to the target terminal.

9. A task processing device based on optoelectronic collaboration, characterized in that, A controller used in a communication system, the device comprising: The first receiving module is used to receive a task request and determine the target data packet corresponding to the task request, wherein the target data packet carries a task tag; The matching module is used to perform feature matching based on the task tag among the service chain templates included in the preset service chain template library to determine the target service chain template. The encapsulation module is used to encapsulate the target data packet based on the service chain identifier of the target service chain template and the task tag to obtain an encapsulated data packet; and to obtain an optoelectronic resource association table based on the encapsulated data packet. The transmission module is used to transmit the encapsulated data packet to the target terminal through the optoelectronic resource association table, optical switch, electrical switch and transmission path, so that the target terminal can perform task processing based on the encapsulated data packet and obtain task results.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images