Method, device, and computer readable storage medium for communication
The AI model-based method for determining composite costs addresses the challenge of handling multiple constraints in networks by optimizing routing and load-balancing, enhancing network performance and efficiency by maximizing constraint-satisfying paths.
Patent Information
- Application Number
- PCT/RU2023/000153
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-26
- Publication Date
- 2024-12-05
AI Technical Summary
Current routing algorithms fail to efficiently handle multiple constraints in networks, leading to intractable problems due to the exponential increase in non-dominated paths, especially in Multi-Objective Shortest Path Problems, where no state-of-the-art solution exists for handling single constraint cases effectively.
A method using an AI model to determine composite costs for network communications by training on network topology, constraints, and metrics, allowing for the consideration of multiple constraints such as delay, bandwidth, and packet loss, and deploying this model in dedicated devices or routers to optimize routing, load-balancing, and scheduling.
This approach enables more accurate and efficient determination of composite costs, maximizing the number of paths satisfying constraints, improving network performance by optimizing routing, load-balancing, and scheduling while reducing computational overhead and ensuring loop-free routing.
Smart Images

Figure RU2023000153_05122024_PF_FP_ABST
Abstract
Description
METHOD, DEVICE, AND COMPUTER READABLE STORAGE MEDIUM FOR COMMUNICATIONFIELD
[0001] Example embodiments of the present disclosure generally relate to the field of telecommunication, and in particular, to a method, a device and a computer readable storage medium for communication.BACKGROUND
[0002] The problem of identifying routes through a network subject to a single constraint is well known. For example, Shortest Path First Algorithms by Dijkstra and Bellman-Ford are well known, and both use a single cost, single constraint approach. Recently, as a consequence of the communication technology development a wide variety of new use cases emerged that the networks must support. Usually, each such use case has different, and multiple requirements for the transmission. For example, some business services are very sensitive to the latency and packet loss, such as remote surgery; while some other services require high bandwidth, such as data transmission.
[0003] The problem of routing while considering multiple constraints has been addressed in a number of ways. It is related to Multi-Objective Shortest Path Problem (MOSPP) and is known to be NP-hard. Hansen proved that, in the worst case, the number of non-dominated paths may increase exponentially with the number of nodes, making this problem intractable in any network of a reasonable scale. Despite tens of years of research, there is still no state- of-the-art algorithm for such problem as Dijkstra for the single constraint case.SUMMARY
[0004] In general, example embodiments of the present disclosure provide a solution for considering multiple constraints when communicating in a network.
[0005] In a first aspect, there is provided a method. The method comprises: providing input data to an Al model, wherein the input data comprises a plurality of metrics of a link in a network; and obtaining, from the Al model, a composite cost corresponding to the link for network communications. In this way, a composite cost for multiple constraints is determined for considering multiple constraints when communicating in a network. This can be achieved1SUBSTITUTE SHEET (RULE 26)by using an Al model to determine a composite cost for network communications. The input data to an Al model is provided, which includes various metrics of a link in a network. The Al model then generates a composite cost corresponding to the link, which takes into account the various constraints in the network. The use of an Al model enables more accurate and efficient determination of composite costs for network communications.
[0006] In some embodiments of the present disclosure, the Al model is obtained based on training data and optimization goal, wherein the training data comprises a topology of the network, a plurality of constraints, and a plurality of metrics of links in the network, and the optimization goal comprises maximizing number of paths satisfying the plurality of the constraints. In this way, the Al model can be trained using training data that includes information about the network topology, the constraints, and the metrics of links in the network. This can further improve the accuracy of the Al model and enable it to maximize the number of paths satisfying the plurality of the constraints, wherein the paths are determined based on a set of composite costs which is generated based the Al model.
[0007] In some embodiments of the present disclosure, the plurality of constraints comprises at least two type of performance of paths in the network satisfies a condition. For example, the plurality of constraints comprises at least two of a delay of a path satisfies a first condition; a bandwidth of the path satisfies a second condition; a packet loss of the path satisfies a third condition; or a jitter of the path satisfies a fourth condition. For example, the delay of the path is required to be less than 20ms, the jitter of the path is required to be less than 5ms, the packet loss of the path is required to be less than 0.005 or the bandwidth of the path is required to be greater than 500Mbps. In this way, the constraints are determined. And by taking into account these constraints, the Al model can generate composite costs that are tailored to specific network requirements.
[0008] In some embodiments of the present disclosure, training the Al model using the training data comprises: generating a plurality of composite costs for all the links based on the plurality of metrics of the links; determining a set of shortest paths to all destinations using the generated plurality of the composite costs; determining whether the generated set of shortest paths reaches the optimization goal; in response to determining that the optimization goal and a stopping criterion are not reached, generating a feedback to update parameters of the Al model or in response to determining that the optimization goal is reached, obtaining the Al model. By this way, the Al model can be trained and the parameters can be updated.2SUBSTITUTE SHEET (RULE 26)
[0009] In some embodiments of the present disclosure, the first path includes the first link and a second link and is determined based on the first composite cost of the first link and a second composite cost of the second link. In this way, the path is determined using plurality of composite costs. The first composite cost of a first link and a second composite cost of a second link can be used to determine a path that satisfies the constraints.
[0010] In some embodiments of the present disclosure, the Al model is deployed in a dedicated device, and providing the input data comprises transmitting the input data from a router to the dedicated device; and obtaining the composite cost comprises receiving the composite cost from the dedicated device. In this way, the Al model can be located on a central device and used to generate composite costs for network communications. The input data is transmitted from a router to the dedicated device, and the composite cost is received back from the dedicated device.
[0011] In some embodiments of the present disclosure, the composite cost is transmitted from the dedicated device to a second router in the network, and then is transmitted from the second router to the router. In this way, the composite cost is distributed to the routers, which can enable more efficient routing of data.
[0012] In some embodiments of the present disclosure, the Al model is deployed in a router, and the method further comprises: receiving, by the router, the Al model from a dedicated device. In this way, the Al model can be located on the router. This can enable more efficient use of resources and reduce the need for dedicated devices.
[0013] In some embodiments of the present disclosure, the method further comprises: distributing, by the router, the Al model to one or more other routers in the network. In this way, the Al model can be distributed to other routers. This can enable more consistent and efficient network communication, as all routers will be using the same Al model for generating composite costs.
[0014] In some embodiments of the present disclosure, the method further comprises: in response to the Al model being retrained at the dedicated device, receiving, by the router, updated parameters of the Al model from the dedicated device; and generating, by the router, an updated Al model based on the updated parameters of the Al model. In this way, the Al model can be updated using the parameters. This can ensure that the Al model remains accurate and effective over time.SUBSTITUTE SHEET (RULE 26)
[0015] In some embodiments of the present disclosure, the method further comprises: receiving, by the router, updated parameters of the Al model from the dedicated device regularly; and generating, by the router, an updated Al model based on the updated parameters of the Al model. In this way, the Al model can be updated regularly. This can ensure that the Al model remains up-to-date with the latest network conditions and constraints.
[0016] In some embodiments of the present disclosure, the method further comprises: determining, by the router, a rout based on a set of composite costs; and routing, by the router, data according to the route. In this way, the route can be determined by the set of composite cost. This can enable more efficient use of network resources and ensure that data is routed in the most optimal way possible.
[0017] In a second aspect, there is provided an apparatus. The apparatus is configured to perform the method of the first aspect or any implementation of the first aspect.
[0018] In a third aspect, there is provided a communication device. The communication device comprising: a communications interface; at least one processor coupled to the communications interface; and a memory coupled to the at least one processor and storing programming instructions for execution by the at least one processor, the programming instructions instruct the at least one processor to: perform the method of the first aspect or any implementation of the first aspect.
[0019] In a fourth aspect, there is provided a computer-readable storage medium having instructions stored thereon that, when executed by one or more processors of a computing device, cause the computing device to perform the method of the first aspect or any implementation of the first aspect.
[0020] It is to be understood that the summary section is not intended to identify key or essential features of embodiments of the present disclosure, nor is it intended to be used to limit the scope of the present disclosure. Other features of the present disclosure will become easily comprehensible through the following description.BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Some example embodiments will now be described with reference to the accompanying drawings, in which:4SUBSTITUTE SHEET (RULE 26)
[0022] FIG. 1 A illustrates an example of a communication system in which some example embodiments of the present disclosure may be implemented;
[0023] FIG. IB illustrates an example of process of training an Al model according to some embodiments of present disclosure;
[0024] FIG. 2A illustrates a first example of architecture of the present disclosure according to some embodiments of present disclosure;
[0025] FIG. 2B illustrates an example of determining the composite costs at dedicated device according to some embodiments of present disclosure;
[0026] FIG. 3A illustrates a second example of architecture of the present disclosure according to some embodiments of present disclosure;
[0027] FIG. 3B illustrates an example of training the Al model at dedicated device according to some embodiments of present disclosure;
[0028] FIG. 3C illustrates an example of determining the composite costs at the router according to some embodiments of present disclosure;
[0029] FIG. 4A illustrates a third example of architecture of the present disclosure according to some embodiments of present disclosure;
[0030] FIG. 4B illustrates a diagram of the Al model structure according to some embodiments of present disclosure;
[0031] FIG. 4C illustrates an example of determining the parameters for Al model at the dedicated device according to some embodiments of present disclosure;
[0032] FIG. 4D illustrates an example of determining the composite costs at the router according to some embodiments of present disclosure;
[0033] FIG. 5 illustrates an example of a method implemented at a device according to some embodiments of the present disclosure; and
[0034] FIG. 6 illustrates a simplified block diagram of a device that is suitable for implementing some example embodiments of the present disclosure.
[0035] Throughout the drawings, the same or similar reference numerals represent the same or similar elements.DETAILED DESCRIPTION5SUBSTITUTE SHEET (RULE 26)
[0036] Principle of the present disclosure will now be described with reference to some example embodiments. It is to be understood that these embodiments are described only for the purpose of illustration and help those skilled in the art to understand and implement the present disclosure, without suggesting any limitation as to the scope of the disclosure. The disclosure described herein can be implemented in various manners other than the ones described below.
[0037] In the following description and claims, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skills in the art to which this disclosure belongs.
[0038] References in the present disclosure to “one embodiment”, “an embodiment”, “an example embodiment”, and the like indicate that the embodiment described may include a particular feature, structure, or characteristic, but it is not necessary that every embodiment includes the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to affect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.
[0039] It shall be understood that although the terms “first” and “second” etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and similarly, a second element could be termed a first element, without departing from the scope of example embodiments. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.
[0040] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises”, “comprising”, “has”, “having”, “includes” and / or “including”, when used herein, specify the presence of stated features, elements, and / or components etc., but do not preclude the presence or addition of one or more other features, elements, components and / or combinations thereof.SUBSTITUTE SHEET (RULE 26)
[0041] As used herein, the term “communication network” refers to a system of interconnected devices and infrastructure that allows for the exchange of information between users. It can be a wireless or wireline network, or a combination of both. A wireless communication network allows for communication over the airwaves without the need for physical wires or cables. This type of network includes technologies such as cellular networks, Wi-Fi, and satellite communication. Wireless networks use electromagnetic waves to transmit data between devices, which can be located anywhere within the network's coverage area. A wireline communication network uses physical cables or wires to transmit data between devices. This type of network includes technologies such as telephone lines, fiber optic cables, and coaxial cables. Wireline networks offer higher speeds and greater reliability than wireless networks, but may be more expensive to install and maintain. In many cases, communication networks use a combination of wireless and wireline technologies to provide connectivity and redundancy. Embodiments of the present disclosure may be applied in various communication networks. Given the rapid development in communications, there will of course also be future type communication technologies and systems with which the present disclosure may be embodied. It should not be seen as limiting the scope of the present disclosure to only the aforementioned networks.
[0042] As used herein, the term “dedicated device” refers to a node in a communication network playing role of a centralized controller. The dedicated device can be a specialized piece of hardware that is designed to perform a specific task or set of tasks within a communication network. In the context of a network controller, a dedicated device is a centralized device that is responsible for managing and controlling the operation of other network devices, such as routers. The dedicated device typically has higher processing power and memory capacity than other devices in the network, allowing it to perform more complex tasks.
[0043] As used herein, the term “router” refers to a network device that is used to connect multiple devices within a network and route data packets between them. Router determines the most efficient path for data to travel between devices, allowing for faster and more reliable communication.
[0044] As mentioned herein before, the problem is how to consider multiple constraints when routing. Routing refers to the process of determining the path for network traffic to take between two points. The goal of routing optimization is to find the most efficient path that, for example, minimizes latency and maximizes throughput, while also considering factors7SUBSTITUTE SHEET (RULE 26)such as network congestion, bandwidth availability, and network topology. It is related to Multi-Objective Shortest Path Problem (MOSPP) and is known to be NP-hard. Hansen proved that, in the worst case, the number of non-dominated paths may increase exponentially with the number of nodes, making this problem intractable in any network of a reasonable scale. Despite tens of years of research, there is still no state-of-the-art algorithm for such problem as Dijkstra for the single constraint case.
[0045] In order to solve this problem, using a composite cost for each link is probably the easiest way to solve the routing problem in a network with multi-metric links. After replacing multiple metrics with a single composite cost, any existing routing solution, for example Dijkstra algorithm, can be used for route computation. In other words, the composite cost is a single-metric that replaces multi-metric (delay, packet loss, jitter, etc.) to compute route or path. Al model, like neural network (NN), can be used to predict the single-metric.
[0046] The present application develops a systematic, scalable and generalized method to synthesize a composite cost to achieve optimal routing plane computation for multiple QoS demands. The present application proposes a new way to synthesize a composite cost as an optimization process maximizing number of paths satisfying constraints in the Shortest Path Tree constructed with the composite cost as an edge weight, which consists of Al model architecture and the training procedure.
[0047] Specifically, the solution of the present application is a routing system and method which use a composite cost in identifying routes. This allows a simple way of identifying the best route taking multiple metrics into account for each link simultaneously. The system allows for the inclusion of multiple constraints, so that the improved method of synthesizing a composite cost for multi-metric links in a network maximizes the number of feasible routes satisfying multiple constraints in the Shortest Path Tree computed using the composite cost. In contrast with existing solutions Al techniques are used to synthesize the composite cost. The method consists of Al model architecture and the training procedure.
[0048] Using the same solution architecture and paradigm, the present disclosure also can achieve the optimization goal of load-balancing. Load-balancing refers to the process of distributing network traffic evenly across multiple network devices or paths, to avoid overloading any one device or path. The goal of load-balancing optimization is to improve network performance and prevent bottlenecks by ensuring that network resources are being utilized efficiently. In this disclosure, the use of Al models to determine composite costs for8SUBSTITUTE SHEET (RULE 26)links in the network can help to balance network traffic across multiple paths, minimizing the overloaded situations for paths.
[0049] The present disclosure also can achieve the optimization goal of scheduling. Scheduling refers to the process of coordinating and prioritizing tasks or processes within a network, to ensure that they are completed in a timely and efficient manner. The goal of scheduling optimization is to improve network performance by ensuring that resources are allocated appropriately, and that tasks are completed in an optimal order. In present disclosure, the use of Al models to determine composite costs for links in the network can help to prioritize network traffic based on the specific constraints and requirements of each task or process, improving overall network performance and efficiency.
[0050] By changing optimization goal, the present disclosure can solve different routing, load-balancing, scheduling problems using the same solution architecture and paradigm.
[0051] FIG. 1 A illustrates an example of a communication system in which some example embodiments of the present disclosure may be implemented. In the communication system, there are a dedicated device 101, and routers 102-110. In some embodiments, the dedicated device 101 can communicate with some of the routers. And one router in the system can communicate with some other routers.
[0052] The scenario of the communication network is distributed routing taking multiple metrics into account for each link simultaneously in a network where a dedicated device is present. The routers 102-110 are connected with each other using a set of links. Each link is described by a set of K metrics, where K >= 2. The values of the link metrics are known to the dedicated device 101 and are communicated by routers 102-110 to dedicated device 101.
[0053] In some embodiments, the input data is provided to an Al model. And the input data comprises a plurality of the metrics of the link in a network. And a composite cost corresponding to the link for network communications is obtained from the Al model. The composite cost is optimized to specific scenario, network topology, metric values characteristics (scale and distribution) and constraints using the same solution architecture and paradigm, and provides systematic approach to synthesizing the optimal composite metric for a specific multi-metric network which guarantees the highest possible discovery of the feasible routes under multiple constraints.
[0054] In some embodiments, the Al model is GNN model. The GNN can be used to design a specific composite cost function for each edge. This leads to higher accuracy of the9SUBSTITUTE SHEET (RULE 26)composite cost and the overall solution as multiple functions are used to synthesize the composite cost for different links.
[0055] The solution of using an optimization approach synthesizes a composite cost which can be used for determining the destinations in the network that can be reached in a tree-like routing structure while satisfying multiple constraints.
[0056] In some embodiments, the dedicated device 101 is used to train the Al model. Training data is provided to the Al model. And the training data comprises a topology of the network, a plurality of constraints, and a plurality of metrics of links in the network. The topology can be a directed graph. The plurality of constraints indicate different goals for network communication. For example, the constraints comprise the values for the metric of communication delay criteria, and the values for the metric of packet loss criteria. For example, the constraints can correspond to the path in the network. For one metric, one value of a constraint can be used for all paths in the network. For example, a delay constraint 20ms is used for all paths in the network, which means the delay of all paths in the network is required to be less than (or less than or equal to) 20ms. And in another way, for one metric, different paths can use different values of the constraint. For example, a delay constraint 20ms is used for some paths in the network, and a delay constraint 5ms is used for other paths in the network. The directed graph is a type of network topology in which the edges have a direction or orientation, meaning that the relationship between the nodes is asymmetrical. In a directed graph, each edge has a source node and a target node, where the source node is the starting point of the edge and the target node is the ending point. This type of topology is commonly used to model systems where there is a flow or direction of information, resources, or control, such as computer networks, transportation systems, and social networks. The directed graph can be used to represent the physical or logical connections between nodes or devices, as well as the direction of data or traffic flow. For example, in a computer network, a directed graph can be used to represent the connections between routers or switches, where each edge represents a physical or logical link between two devices, and the direction of the edge indicates the direction of data flow.
[0057] In present disclosure, the directed graph topology is used to model the network, where each node represents a device in the network, and each edge represents a link between two devices. The direction of the edge indicates the direction of data flow, and the Al model is used to determine the composite cost of each link based on multiple constraints, such as10SUBSTITUTE SHEET (RULE 26)delay, bandwidth, and packet loss, in order to optimize routing, load balancing, or scheduling in the network.
[0058] In some embodiments of the present disclosure, the plurality of constraints comprises at least two type of performance of paths in the network satisfies a condition.
[0059] In some embodiments, training the Al model using the training data may include following steps: generating a plurality of composite costs for all the links based on the plurality of metrics of the links; determining a set of shortest paths to all destinations using the generated plurality of composite costs; determining whether the generated shortest path reaches an optimization goal based on the plurality of constraints; in response to determining that the optimization goal and a stopping criterion are not reached, generating a feedback to update parameters of the Al model. The optimization goal comprises maximizing number of paths satisfying the plurality of constraints. In some embodiments, paths includes all connections between any two nodes in the network. In some embodiments, paths includes part of the all connections. For example, paths only include connections between nodes which is related to one or several services. For another example, paths only include connections between nodes which is specified by a user. In some embodiments, if all paths of a training satisfy the plurality of constraints, the optimization goal is reached. In some embodiments, if a quantity of paths satisfying the plurality of constraints exceeds a quantity threshold, the optimization goal is reached. In some embodiments, if a proportion of paths satisfying the plurality of constraints exceeds a proportion threshold, the optimization goal is reached. If the optimization goal is reached, the training can stop, and current Al model is used as a trained model. If the optimization goal is not reached and the termination condition is reached, the best result in the training procedure can be output. The result of a training can be recorded. A result of each training includes, for example, a quantity or a proportion of paths that are obtained based on the training and that meet constraints. The best result, for example, is the result with maximum quantity or maximum proportion. For example, results of each training are recorded, and the best result is selected by comparing the results when the termination condition is reached. For example, a new result is compared to a recorded result, and the recorded result is replaced by the new result if the new result is better than the recorded result. In other words, only one result need to be recorded, the best result is the recorded result when the termination condition is reached. Model parameters corresponding to the best result can also be recorded. When the best result is outputted, the feedback from a user (e.g., operation11SUBSTITUTE SHEET (RULE 26)and maintenance personnel) can be received. If the user accepts the best result of feedback, the recorded model parameters corresponding to the best result can be directly used.
[0060] FIG. IB illustrates an example of how to train the Al model. First, the input of training data is provided to the Al model. The training data includes information about the topology of the network, a set of constraints, and metrics of links in the network. The purpose of training the Al model is to enable it to accurately determine the composite cost of links in the network while considering multiple constraints. For example, the topology in FIG. 1 A is provide to the Al model. And the constraints of link between all routers in FIG. 1A are provided to the Al model. And the metrics of link between all routers in FIG.1A also are provided to Al model.
[0061] Then at 120, the process of training the Al model includes that a plurality of composite costs for all the links is generated based on the plurality of metrics of the links. For example, the composite costs between all routers in Fig. 1 A are generated based on the input.
[0062] At 121, a set of shortest paths to all destinations are determined using the plurality of the composite costs. For example, there is a plurality of routing destinations in Fig. 1A. Then all the shortest paths for the destinations are determined based on the composite costs. The current Al model is used to generate the composite costs for all the links in the network.
[0063] At 122, the result of whether the determined shortest paths reaches an optimization goal is generated. The optimization goal comprises maximizing number of paths that satisfy the plurality of the constraints. Paths are determined based on the plurality of the composite costs. For example, if all the paths satisfy the plurality of constraints, the optimization goal is reached. For example, if the quantity of the paths satisfying the plurality of constraints exceeds the quantity threshold, the optimization goal is reached. For example, if the proportion of the paths satisfying the plurality of constraints exceeds the proportion threshold, the optimization goal is reached.
[0064] At 123, in response to determining that the optimization goal and a stopping criterion are not reached, a feedback is generated to update parameters of the Al model. For example, the paths do not meet the optimization goal and the stopping criterion of training the Al is not reached, then the feedback for updating the Al model is generated and input to the Al model. For example, reaching optimization goal is also a stopping criterion.12SUBSTITUTE SHEET (RULE 26)
[0065] By providing feedback to the Al model during the training process, the model can learn from its mistakes and adjust its parameters based on the differences between the target value and the current phased output value, to better satisfy the plurality of constraints when generating composite costs for links in the network.
[0066] At 124, if the optimization goal is not reached and the termination condition is reached, the best result in the training procedure can be output, and a proportion of the best result or the quantity of the best result can be output. The feedback from the user can be received. If the user accepts the best result, the recorded parameters corresponding to the best result can be directly used and the trained Al model is obtained.
[0067] In some embodiments, after training is completed, the Al model (the trained Al model) is used to compute composite costs for links in the network. Composite costs are used to determine paths in the network. In some embodiments, the paths includes a first path, the first path includes a first link and a second link. And the first path is determined based on a first composite cost of the first link and a second composite cost of the second link. A composite cost is obtained by providing input data to the Al model, which generates a composite cost corresponding to the link for network communications. In other words, the Al model determines the composite costs of each link in the network based on the input data that includes a plurality of metrics of links in the network.
[0068] Thus, the invention provides an efficient method of determining the best path for network communications by taking into account multiple constraints and using an Al model to generate composite costs for links in the network.
[0069] FIG. 2A illustrates a first example of architecture of the present disclosure according to some embodiments. According to FIG 2A, the Al model is deployed in the dedicated device 101. And the input data is transmitted from routers 102-110 to the dedicated device 101. And the routers 102-110 receive composite costs from the dedicated device 101.
[0070] In some embodiments, when requirement to generate the composite costs by the Al model exceeds routers 102-110 capabilities or network performance requirements, the composite costs are generated in the dedicated device 101 and then are communicated to the routers 102-110.
[0071] In some embodiments, the dedicated device 101 is used to train an Al model generating the composite cost and to inference the composite costs for all the connections. The dedicated device 101 also communicates the inferenced composite cost values to routers13SUBSTITUTE SHEET (RULE 26)102-110 for making routing decisions. For example, as shown in FIG. 2A, the dedicated device 101 communicate (201) the inferenced composite cost values to routers 105. Then any algorithm using single constraint approach, for example, Shortest Path First Algorithm by Dijkstra or Bellman-Ford is used to identify routes. In this way, using a composite cost for route identification with a Shortest Path First Algorithm guarantees that the routing is loop- free and fully distributed. The solution is applicable to general topology and any graph scale.
[0072] The present disclosure solves the problem of high computation overhead for an Al model (e.g., NN) training, while keeping the ability to use the composite cost for route computation in a commodity hardware. The solution is applicable in multiple scenarios with a dedicated device in the network. The present disclosure solves the problem of fast MOSPP solving (for multiple destinations at once) which is an NP-hard problem by splitting the timeconsuming part of NN training from routing path computation in a router which is done using any efficient SPF algorithm. The present disclosure also provides compatibility with current router architecture, low resource requirements and high execution speed with obligatory guarantees for loop-free routing in distributed mode.
[0073] FIG. 2B illustrates an example of determining composite costs at dedicated device according to some embodiments of present disclosure. In some embodiments, the Al model is deployed in the dedicated device 101, and the routers 102-110 transmit the input data to the dedicated device 101. And the dedicated device 101 then provides the composite costs using the Al model.
[0074] According to FIG. 2B, at dedicated device 101, training data 202 is provided to training block 203. The training data is used in the training block 203 to train the Al model. Then the output of the training block 203 is the trained model 204. The trained model is used for generating composite cost for each link in the network. And the output of the dedicated device 101 is the set of composite costs 205 for all the edges. Then these composite costs are communicated 206 to each router in the network.
[0075] FIG. 3A illustrates a second example of architecture of the present disclosure according to some embodiments. In some embodiments, the Al model is deployed in the routers 102-110. The routers 102-110 receive the Al model from the dedicated device 101. For example, as shown in FIG. 3A, the router 105 receives (301) the Al model from the dedicated device 101.SUBSTITUTE SHEET (RULE 26)
[0076] In some embodiments, when execution time and size of the Al model are reasonable, the Al model is communicated by the dedicated device 101 to the routers 102-110, and then the routers 102-110 generate the composite costs themselves for further route identification.
[0077] FIG. 3B illustrates an example of training the Al model at dedicated device according to some embodiments. According to FIG. 3B, at dedicated device 101, training data 302 is provided to training block 303. The training data is used in the training block 303 to train the Al model for generating the composite cost. Then the output of the training block 303 is the trained model 204 which will be sent 305 to the routers 102-110.
[0078] In this way, while the architecture of the overall routing system is a combination of distributed mode (for routing) with the centralized mode (for the Al model training), when the Al model is trained, the route computation is fully distributed, so the solution can be easily deployed in the current router architecture. The Al model represents a dynamic nature of the composite cost, which allows dynamic adaptation to the changing network conditions. A NN as a representation of the composite cost as a function can be updated in contrast to static composite cost functions.
[0079] FIG. 3C illustrates an example of determining the composite costs at the router according to some embodiments. According to FIG. 3C, at router 102, trained model 307 is received 308 from the dedicated device 101. Then the input data 306 is provided to the trained model 307. The trained model 307 inferences the composite cost and outputs a set of composite cost values 309 for each edge. Then the composite cost values 309 are used 310 for computing shortest path. For example, when all the link composite costs are inferenced in router 102, any algorithm using single constraint approach, for example, Shortest Path First Algorithm by Dijkstra or Bellman-Ford are used to identify routes by the composite costs. And the Shortest Path First algorithm is pre-installed in the router 102.
[0080] FIG. 4A illustrates a third example of architecture of the present disclosure according to some embodiments. In some embodiments, when the Al model is retrained at the dedicated device 101, the routers 102-110 receive updated parameters of the Al model from the dedicated device 101. And then the routers 102-110 generate the updated Al model based on the updated parameters of the Al model. For example, the router 105 receives (401) the Al model. And the Al model is retrained at the dedicated device 101 and the updated parameters for the Al model are generated. The router 105 receives (402) the updated parameters from the dedicated device 101. In some embodiments, the dedicated device 10115SUBSTITUTE SHEET (RULE 26)communicates once the initial trained model (or periodically - the full updated model) to all other devices (routers) making routing decisions. And the dedicated device 101 is to communicate regularly the parameters of the trained / updated model to all other devices (routers) making routing decisions.
[0081] In large-scale network updating NN model can lead to significant communication overhead. The present disclosure is an efficient architecture to decrease the communication to literally a few dozens of bytes per a router.
[0082] FIG. 4B illustrates a diagram of the Al model structure according to some embodiments of present disclosure. In some embodiments, when the Al model is updated and the architecture of the Al model stays the same, only the model parameters are communicated to the routers 102-110.
[0083] FIG. 4C illustrates an example of determining the parameters for Al model at the dedicated device according to some embodiments of present disclosure. In some embodiments, when an Al model size is big or communicating the model exceeds available bandwidth, only the necessary model parameters can be communicated by the dedicated device to the routers to decrease the communication overhead, and then the routers synthesize the composite costs.
[0084] According to FIG. 4C, the dedicated device 101 is used to train an Al model synthesizing a composite cost. In some embodiments, at dedicated device 101, training data 402 is provided to training block 403. The training data is used in the training block 403 to train the Al model for synthesizing the composite cost. The output of the training block 403 is the trained model 404 to be updated in the routers 102-110. Then according to the trained model 404, the set of model parameters 405 is determined and transmitted 406 to routers 102- 110.
[0085] FIG. 4D illustrates an example of determining the composite costs at the router according to some embodiments. According to FIG. 4D, at router 102, trained model 409 and updated parameters are received 408 from the dedicated device 101. The trained model 409 is updated using the updated parameters. The input data 407 is provided to the trained model 409. The trained model 409 inferences the composite cost and outputs a set of composite cost values 410 for each edge. Then the composite cost values 410 are used 411 for computing shortest path. For example, when all the link composite costs are inferenced in router 102, any algorithm using single constraint approach, for example, Shortest Path First16SUBSTITUTE SHEET (RULE 26)Algorithm by Dijkstra or Bellman-Ford is used to identify routes by the composite costs. And the Shortest Path First algorithm is pre-installed in the router 102.
[0086] In some embodiments, the routing system 100 is a Space-Ground Integrated Network (SGIN) which consists of one (or many) stationary Gateways and a massive constellation of LEO-satellites. Each LEO-satellite exploits a fast and efficient single-cost SPF-algorithm to identify routes. LEO-satellites periodically communicate the connection link parameters to a Gateway. The gateway (or multiple gateways) collect the necessary information and train the Al model to inference composite costs. Then the composite costs are communicated to LEO-satellite. Finally, LEO-satellite uses its’ fast and efficient SPF- algorithm with the composite costs to identify routes.
[0087] In some embodiments, the routing system 100 is a Space-Ground Integrated Network (SGIN) which consists of one (or many) stationary Gateways and a massive Multilayer (LEO / MEO / GEO) satellite constellation. The constellation itself consists of LEO, MEO and GEO satellites. LEO-satellites periodically communicate the connection link parameters to a Gateway and corresponding MEO-satellite. The gateway (or multiple gateways) collect the necessary information and train the Al model to inference composite costs. Then the neural network is communicated to GEO -satellite. MEO-satellites aggregate link parameters and communicate them to GEO-satellite. GEO-satellite, using the NN model, inference composite costs and use them to identify the routes. Then these routes are communicated to MEO-satellites, and MEO-satellites communicate the routes to LEO- satellites.
[0088] FIG. 5 illustrates an example of a method implemented at a device in according to some embodiments. At block 510, input data is provided to an Al model and the input data comprises a plurality of metrics of a link in a network. At block 520, a composite cost corresponding to the link for network communications is obtained from the Al model.
[0089] In some embodiments, the method further comprises: providing training data to the Al model, wherein the training data comprises a topology of the network, a plurality of constraints, and a plurality of metrics of links in the network; and training the Al model using the training data. This way, the Al model can be trained.
[0090] In some embodiments of the present disclosure, the plurality of constraints comprises at least two type of performance of paths in the network satisfies a condition.17SUBSTITUTE SHEET (RULE 26)
[0091] In some embodiments, the Al model is obtained based on training data and optimization goal. The training data comprises a topology of a network, a plurality of constraints, and a plurality of metrics of links in the network. The optimization goal comprises maximizing number of paths satisfying the plurality of the constraints.
[0092] In some embodiments, training the Al model using the training data comprises: generating a plurality of composite costs for all the links based on the plurality of metrics of the links; determining a set of shortest paths to all destinations using the generated plurality of composite costs; determining whether the determined shortest paths reaches the optimization goal; in response to determining that the optimization goal and a stopping criterion are not reached, generating a feedback to update parameters of the Al mode; or in response to determining that the optimization goal is reached, obtaining the Al model; or in response to determining that the optimization goal is not reached and the stopping criterion is reached, outputting the best result of the best result in the training procedure, and in response to receiving an acceptation feedback, determining an Al model corresponding to the best result as the Al model (the trained Al model).
[0093] In some embodiments, the Al model is deployed in a dedicated device, and providing the input data comprises transmitting the input data from a router to the dedicated device. And the routers receive the composite costs from the dedicated device. This way, the Al model can be located on a central device.
[0094] In some embodiments, the composite costs are transmitted from the dedicated device to a part of routers in the network, and then is transmitted from the part of routers to other routers. This way, the composite costs is distributed to the routers. The dedicated device can send relevant information (e.g., Al model, Al model parameters, composite costs, etc.) to each router separately. The controller can also send relevant information to only a part of the routers (the dedicated device can send the routers all or part of the information), and then the router that receives the information sends relevant information (e.g., based on IGP flooding mechanism) to other routers so that all routers receive it.
[0095] In some embodiments, the Al model is deployed in a router, and the method further comprises: receiving, by the router, the Al model from a dedicated device. This way, the Al model can be located on the router.SUBSTITUTE SHEET (RULE 26)
[0096] In some embodiments, the method further comprises: distributing, by the router, the Al model to one or more other routers in the network. This way, the Al model can be distributed to other routers.
[0097] In some embodiments, the method further comprises: in response the Al model being retrained at the dedicated device, receiving, by the router, updated parameters of the Al model from the dedicated device; and generating, by the router, an updated Al model based on the updated parameters of the Al model. This way, the Al model can be updated using the parameters.
[0098] In some embodiments, the method further comprises: receiving, by the router, updated parameters of the Al model from the dedicated device regularly; and generating, by the router, an updated Al model based on the updated parameters of the Al model. This way, the Al model can be updated regularly.
[0099] In some embodiments, the method further comprises: determining, by the router, a route based on a set of composite costs; and routing, by the router, data according to the route. This way, the route can be determined by the set of composite cost.
[0100] FIG. 6 illustrates a simplified block diagram of a device 600 that is suitable for implementing some example embodiments of the present disclosure. For example, the device 600 may be provided to implement the dedicated device 101, the routers 102-110 as shown in FIG. 1 A.
[0101] As shown in FIG. 6, the communication device 600 includes one or more processors 610, one or more memories 630, and one or more communication interfaces 620. The processor 610 may be of any type suitable to the local technical network and may include one or more of the following: general purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs) and processors based on multicore processor architecture, as non-limiting examples. The communication device 1100 may have multiple processors, such as an application specific integrated circuit chip that is slaved in time to a clock which synchronizes the main processor.
[0102] The memory 630 may include one or more non-volatile memories and one or more volatile memories. Examples of the non-volatile memories include, but are not limited to, a Read Only Memory (ROM), an electrically programmable read only memory (EPROM), a flash memory, a hard disk, a compact disc (CD), a digital video disk (DVD), and other magnetic storage and / or optical storage. Examples of the volatile memories include, but are19SUBSTITUTE SHEET (RULE 26)not limited to, a random access memory (RAM) and other volatile memories that will not last in the power-down duration.
[0103] The communication interface 620 can be used for bidirectional communications. The communication interface 620 may have at least one antenna to facilitate communication. The communication interface 620 may represent any interface that is necessary for communication with other network elements.
[0104] The processor 610 is configured to control the communication interface 620 to receive and send a signal. The memory 630 is configured to store a computer program. The processor 610 is configured to invoke the computer program from the memory 630 and run the computer program, so that the communication device 610 is enabled to perform a corresponding procedure and / or operation in various embodiments of the communication methods in this application.
[0105] A person of ordinary skill in the art may be aware that, in combination with units and algorithm steps of the examples described in the embodiments disclosed in this specification, this application may be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed by hardware or software depends on particular applications and design constraint conditions of the technical solutions. A person skilled in the art may use different methods to implement the described functions for each particular application, but it should not be considered that the implementation goes beyond the scope of this application.
[0106] It may be clearly understood by a person skilled in the art that, for convenient and brief description, for a detailed working process of the foregoing system, apparatus, and unit, refer to a corresponding process in the foregoing method embodiments, and details are not described herein again.
[0107] In the several embodiments provided in this application, it should be understood that the disclosed system, apparatus, and method may be implemented in other manners. For example, the described apparatus embodiment is merely an example. For example, the unit division is merely logical function division and may be other division in actual implementation. For example, a plurality of units or components may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections may be implemented by using some interfaces. The indirect couplings or20SUBSTITUTE SHEET (RULE 26)communication connections between the apparatuses or units may be implemented in electronic, mechanical, or other forms.
[0108] The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of network units. Some or all of the units may be selected based on actual requirements to achieve the objectives of the solutions of the embodiments.
[0109] In addition, functional units in the embodiments of this application may be integrated into one processing unit, or each of the units may exist alone physically, or two or more units are integrated into one unit.
[0110] When the functions are implemented in the form of a software functional unit and sold or used as an independent product, the functions may be stored in a computer-readable storage medium. Based on such an understanding, the technical solutions of this application essentially, or the part contributing to the prior art, or some of the technical solutions may be implemented in a form of a software product. The software product is stored in a storage medium, and includes several instructions for instructing a computer device (which may be a personal computer, a server, or a network device) to perform all or some of the steps of the methods described in the embodiments of this application. The foregoing storage medium includes: any medium that can store program code, such as a USB flash drive, a removable hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk, or an optical disc.
[0111] The foregoing descriptions are merely specific implementations of this application, but are not intended to limit the protection scope of this application. Any variation or replacement readily figured out by a person skilled in the art within the technical scope disclosed in this application shall fall within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claims.SUBSTITUTE SHEET (RULE 26)
Claims
WHAT IS CLAIMED IS:
1. A method comprising: providing input data to an Al model, wherein the input data comprises a plurality of metrics of a link in a network; and obtaining, from the Al model, a composite cost corresponding to the link for network communications.
2. The method of claim 1, wherein the Al model is obtained based on training data and optimization goal; wherein the training data comprises a topology of the network, a plurality of constraints, and a plurality of metrics of links in the network; and the optimization goal comprises maximizing number of paths satisfying the plurality of the constraints.
3. The method of claim 2, wherein the plurality of constraints comprises at least two type of performance of paths in the network satisfies a condition.
4. The method of claim 2 or 3, wherein the method further comprises training the Al model based on the training data and the optimization goal, and wherein the training the Al model comprises: generating a plurality of composite costs for all the links based on the plurality of metrics of the links; determining a set of shortest paths to all destinations using the generated plurality of composite costs; determining whether the determined set of shortest paths reaches the optimization goal; in response to determining that the optimization goal and a stopping criterion are not reached, generating a feedback to update parameters of the Al model; or in response to determining that the optimization goal is reached, obtaining the Al model.
5. The method of any one of claims 1 to 4, wherein the method further comprises:22SUBSTITUTE SHEET (RULE 26)determining a first path, wherein the first path includes a first link and a second link and the first path is determined based on a first composite cost of the first link and a second composite cost of the second link.
6. The method of any one of claims 1-5, wherein the Al model is deployed in a dedicated device, and providing the input data comprises transmitting the input data from a router to the dedicated device; and obtaining the composite cost comprises receiving the composite cost from the dedicated device.
7. The method of claim 6, wherein the composite cost is transmitted from the dedicated device to a second router in the network, and then is transmitted from the second router to the router.
8. The method of any one of claims 1-5, wherein the Al model is deployed in a router, and the method further comprises: receiving, by the router, the Al model from a dedicated device.
9. The method of claim 8, further comprising: distributing, by the router, the Al model to one or more other routers in the network.
10. The method of claim 8, further comprising: in response the Al model being retrained at the dedicated device, receiving, by the router, updated parameters of the Al model from the dedicated device; and generating, by the router, an updated Al model based on the updated parameters of the Al model.
11. The method of claim 8, further comprising: receiving, by the router, updated parameters of the Al model from the dedicated device regularly; and generating, by the router, an updated Al model based on the updated parameters of the Al model.23SUBSTITUTE SHEET (RULE 26)12. The method of any one of claims 1-11, further comprising: determining, by a router, a route based on a set of composite costs, wherein the set of composite costs are generated by the Al model and are corresponding to a plurality of links in the network; and routing, by the router, data according to the route.
13. An apparatus, wherein the apparatus is configure to perform the method of any one of claims 1-12.
14. A device comprising: a communications interface; at least one processor coupled to the communications interface; and a memory coupled to the at least one processor and storing programming instructions for execution by the at least one processor, the programming instructions instruct the at least one processor to perform the method of any one of claims 1-12.
15. A computer-readable storage medium having instructions stored thereon that, when executed by one or more processors of a device, cause the device to perform the method of any one of claims 1-12.SUBSTITUTE SHEET (RULE 26)
Citation Information
Patent Citations
Multi-constraint routing system and method
US20030118024A1
Predictive overlay network architecture
US20180309636A1
Replacing static routing metrics with probabilistic models
US20220272029A1
Cited By
Multi-satellite cooperative distributed routing method for multi-agent reinforcement learning
CN119995693A
