A method for joint recognition and scheduling transmission of non-homogeneous information

Abstract

The present invention provides a method for joint recognition and scheduling transmission of heterogeneous information. By integrating optical converters into the DCN topology, the network is given the inherent ability to self-adjust according to traffic patterns. Compared with existing layered or extended graph solutions, the present invention has the ability to switch network topologies more flexibly, even when data center services are in progress.

Description

Technical Field

[0001] The present invention belongs to the technical field of network communication and data processing, and in particular relates to a method for joint recognition and scheduling transmission of non-homogeneous information. Background Art

[0002] Data centers widely adopt topologies such as CLOS, BCube, and Xpander to manage their vast device networks. Data center networks (DCNs) rely on traditional topologies that are inherently static and lack flexibility to changes in traffic patterns. With technological advancements, researchers have recently designed dynamic DCN topologies based on multimodal converters, providing an efficient solution without expanding existing communication resources. However, fully leveraging the heterogeneous information in DCNs and achieving network-wide joint optimization of converters remains a major challenge in algorithm design. This patent applies graph representation learning algorithms to DCN topology optimization for the first time and proposes an innovative solution, MorpheusClos.

[0003] In recent years, data center networks (DCNs) have developed rapidly to meet the growing demands of emerging applications. The new applications they support, such as cloud computing and large-scale data learning, require more powerful network communication technologies. These emerging applications often require more intensive data exchange and parallel processing services. Notably, these services often change over time. Traditional DCNs, on the other hand, rely on electrical connections to coordinate switch devices to establish a fixed topology. However, fixed topologies cannot meet the topology-aware needs of emerging applications and are unable to respond to changing traffic patterns in the network. As a recent research achievement, optical switches with adjustable connections provide a groundbreaking solution for design, enabling the construction of network topologies that truly adapt to demand.

[0004] Unlike traditional electrical connections, these converters provide a more flexible solution for switch connections. Figure 1The top half of the figure shows three modes of a six-port switch, which implements port switching via its physical optical splitters. Before the advent of these switches, the development of DCN topology design faced two conflicting goals: ease of deployment and good performance. Industry leaders such as Google and Alibaba chose to adopt the multi-level Clos architecture to build their DCNs, reflecting the industry's de facto preference. Clos, with its tree-like structure based on device clusters (pods), has long been considered the gold standard for DCN topology design. Its advantages lie in its ability to deploy the system using simple rules and its inherent scalability with the introduction of additional servers. However, Clos's multi-layered structure increases the number of routing hops between any two servers, reducing network efficiency in terms of throughput and latency. In contrast, academics have introduced flatter extended graph architectures such as Jellyfish and Xpander, which reduce the number of routing hops between servers by randomly interconnecting them. A recent study demonstrated that extended architectures can achieve similar throughput performance to Clos while using 25% fewer switches. However, these flatter architectures also increase complexity. Even worse, the scalable graph framework introduces the possibility of routing loops in its topology, which can cause priority-based flow control (PFC) deadlocks. Therefore, when considering the performance benefits of scalable graphs, the risk of PFC deadlocks must also be considered.

[0005] Given these conflicting aspects, many researchers have strived to find a balance between the rigid hierarchical structure of CLOS-type architectures and the randomly hierarchical, flat configurations of extended graph-type designs. Significant work in this area includes Diamond, SlimFly, and FatClique, but a widely accepted general solution has yet to be found. This is because time-varying traffic patterns have preferred network topologies, and a dynamic balance between hierarchical and flat solutions is difficult to achieve. As mentioned earlier, the use of tunable switches presents a potential solution. Figure 1 The lower half of the figure shows the CLOS architecture before and after the integration of the converters. The dotted lines represent possible connections, but due to the inherent limitations of physical optical splitters, the optical converters cannot be connected to all ports simultaneously. These converters replace electrical wiring in the network, and the goal of this patent is to enable the DCN topology to dynamically adapt to network traffic patterns, allowing the network topology to be reshaped in real time based on traffic demand. However, a key question remains: how can these converters adapt to the ever-changing network traffic? As data volumes surge and networks expand, the ability of these converters to accurately read and respond to various types of traffic becomes critical. Summary of the Invention

[0006] To solve the above problems, the present invention provides a method for joint recognition and scheduling transmission of non-homogeneous information, which can switch network topology more flexibly.

[0007] A method for joint recognition and scheduling of heterogeneous information transmission is disclosed. Traffic is transmitted from a sending server to a receiving server via a meta-path composed of switches located at different layers of a heterogeneous data center network. The switches on the heterogeneous data center network are divided into core layer switches, aggregation layer switches, and edge layer switches. Adjacent switches on the meta-path are connected by converters, and the number of converters is the smaller of the number of aggregation layer switches and edge layer switches. Different operating modes of the converters correspond to different switch connection methods, and the operating mode of the converters changes with the traffic transmission requirements of the heterogeneous data center network. When the traffic transmission meta-path of the heterogeneous data center network needs to be changed in the next time slot, the method for obtaining the operating mode of each converter in the next time slot is as follows:

[0008] S1: Obtaining status information of a heterogeneous data center network in a current time slot, wherein the heterogeneous data center network is composed of multiple converters, multiple switches, and multiple servers, and the status information includes service control information and topological location information of each converter in a current time slot, service control information and topological location information of each server in a current time slot, and topological location information and dynamic operation information of each switch in a current time slot;

[0009] S2: Input the state information into the trained decision system to obtain the optimal working mode of each converter in the next time slot.

[0010] Furthermore, in step S2, the different operating modes of the switches, servers, and converters in the heterogeneous data center network are used as nodes of the heterogeneous data center network topology; at the same time, the decision system includes multiple cascaded Dyta modules, multi-layer perceptrons (MLPs), and PUCB models, wherein each Dyta module consists of a Dyta layer and a causal convolutional layer;

[0011] Except for the Dyta layer of the first-level Dyta module, the Dyta layers of the remaining Dyta modules are used to process the first node feature information stored in the form of a heterogeneous data center network topology structure obtained by the previous-level Dyta module based on the HDCN mutual attention mechanism to obtain second node feature information; the causal convolution layer is used to fuse the second node feature information obtained by the Dyta layer of the same Dyta module with the second node characteristic information of different time slots in history to obtain first node feature information used to describe the causal relationship between topology scheduling and network performance; wherein, the Dyta layer of the first-level Dyta module is used to convert state information into second node feature information based on the HDCN mutual attention mechanism, and the first node feature information obtained by the causal convolution layer of the last-level Dyta module is output to the multi-layer perceptron MLP;

[0012] The multilayer perceptron MLP is used to obtain the prior probability that each operating mode of each converter in the next time slot is the optimal operating mode based on the first node feature information output by the previous stage Dyta module;

[0013] The PUCB model is used to calculate the corresponding upper confidence bounds of each converter when it is in various working modes in the next time slot based on the corresponding prior probabilities of each converter when it is in various working modes in the next time slot and the percentage of the remaining link capacity of each converter in the previous time slot, and to use the working mode with the maximum upper confidence bound value among multiple modes of each converter as the working mode of each converter in the next time slot.

[0014] Furthermore, the second node feature information obtained by the Dyta layer of any level Dyta module is composed of the target node output features corresponding to each node in the network, and the target node output features corresponding to any network node are obtained as follows:

[0015] Take each network node as the target node and perform the following steps to obtain the updated target node output features corresponding to each network node:

[0016] S21: Get the target node v t The adjacent nodes and the network nodes with similar traffic patterns in the previous time slot are used as source nodes v s ;

[0017] S22: Each source node v s The service control information and topological location information of the previous time slot are used as the original features of the source node, and the original features of the source node are reduced to obtain the source node input feature H (l-1) [v s ], where l represents the source node v s The hierarchical number of the Dyta module where it is located; the source node input feature of the first-level Dyta module is defined as H0 [v s ];

[0018] S23: Set each target node v t The service control information and topological location information of the previous time slot are used as the original features of the target node, and the original features of each target node are reduced to obtain the target node input feature H (l-1) [v t ]; the target node input feature of the first-level Dyta module is defined as H 0 [v t ];

[0019] S24: Input each target node into the feature H through the query vector linear projection function Q-linear l-1 [v t ] are mapped to multiple vector spaces respectively, and the query vector mapping Q of each vector space is obtained h (v t ):

[0020] Q h (v t )=Q-linear h (H (l-1) [v t ])

[0021] Wherein, the vector space serial number h=1,2,...,H, H represents the number of vector spaces;

[0022] S25: Input the source node into the feature H through the key vector linear projection function K-linear (l-1) [v s ] is mapped to multiple vector spaces to obtain the key vector mapping K of each vector space h (v s ):

[0023] K h (v s )=K-linear h (H (l-1) [v s ])

[0024] S26: Input the source node into the feature H through the value vector linear projection function V-linear l [v s ] is mapped to multiple vector spaces to obtain the value vector mapping V of each vector space h (v s ):

[0025] V h (v s )=V-linear h(H (l-1) [v s ])

[0026] S26: Map K according to the key vector h (v s ) and query vector mapping Q h (v t ) Get the source node v s For each target node v t The importance of Attention(v s ,v t ):

[0027]

[0028] Among them, ATT_head h (v s ,v t ) represents any target node v in the hth vector space t and source node v s The attention weight between is calculated as:

[0029]

[0030] Where d represents the target node input feature H (l-1) [v t ] dimension, Indicates the use to capture the source node v s and target node v t The transformation matrix of the characteristic relationship between Represents the source node v s and target node v t The attention adaptive scaling factor between , T represents transposition;

[0031] S27: Using linear projection Map the value vector V h (v s ) is converted to a value message ValueMessage(v s ,v t ):

[0032]

[0033] Among them, MSG-head h (v s ,v t ) indicates that the set transformation matrix is ​​used Mapping V with value vector h (v s ) multiplied by, Indicates that all vector spaces correspond to the MSG-head h (v s ,v t ) added;

[0034] S28: The importance of Attention (v s ,v t ) and value message ValueMessage(v s ,v t ) to obtain each target node v t Message Vector

[0035]

[0036] Where SN(t) represents the set of source nodes selected in the current time slot t, It represents the product Attention(v) corresponding to the node pairs consisting of all source nodes and their paired target nodes in the set SN(t). s ,v t )·ValueMessage(v s ,v t ) added;

[0037] S29: Get the target node v according to the nonlinear activation ReLu function t The target node output feature H is output at the current level Dyta module (l) [t]:

[0038]

[0039] Among them, σ is the weight of the nonlinear activation ReLu function, H (l-1) [t] is the target node v t Output features at the target node output by the previous level Dyta module.

[0040] Furthermore, the method for the causal convolution layer of any level Dyta module to obtain the feature information of the first node is as follows:

[0041]

[0042] in, is the output feature of the rth causal convolution of the current causal convolution layer at the current time slot t, is the first node feature information finally obtained by the current causal convolution layer at the current time slot t, D is the expansion factor used to characterize the number of layers in the causal convolution layer, K is the number of convolutions performed in each layer of causal convolution, is the weight of the rth causal convolution layer in the current lth level Dyta module when performing the i-th convolution, G t-i×r represents the graph data within the expansion time step, where the graph data is obtained by treating the different working modes of each switch, server, and converter as a node to form a topology graph; the state information of the K×D topology graphs of the historical K×D time slots including the current time slot t is used as the second node feature information obtained in the current l-th level Dyta module as the graph data; σ represents the activation function, and ReLU represents the nonlinear activation ReLu function.

[0043] Furthermore, the method for determining the optimal operating mode of each converter in the next time slot is as follows:

[0044] Construct a multi-level search tree for the PUCB model, wherein the last-level leaf node of the search tree corresponds to the next time slot, and each of the remaining nodes corresponds to a historical time slot. Each node represents a converter operating mode and its related statistical information, wherein the statistical information includes the number of times it was selected as the optimal operating mode, the percentage of remaining link capacity in the previous time slot, and the prior probability. At the same time, except for the root node and leaf nodes, the remaining nodes serve as both parent nodes and child nodes.

[0045] The search tree is traversed according to the set search rules, and the confidence upper bounds of the converter operating modes corresponding to the nodes at each level are calculated level by level. The set search rules are that the nodes with fewer selections are selected first, and the nodes with larger confidence upper bounds are selected first. At the same time, the confidence upper bound of the converter operating mode corresponding to any node is obtained as follows:

[0046]

[0047] Where U(v,t) represents the confidence upper bound of the converter operating mode corresponding to the current node v in the time slot t corresponding to its own level, N(v,t) is the number of times the converter operating mode corresponding to the current node v is selected as the optimal operating mode before the end of the time slot t corresponding to its own level, P(v,t) is the highest a priori probability of the converter operating mode corresponding to the current node v in the time slot t corresponding to its own level, c puct is a setting constant that balances exploration and exploitation;

[0048] For each parent node, select the child node with the largest Q(v,t)+U(v,t) from its respective child nodes as the optimal child node:

[0049] Optimal child node ← argmax v (Q(v,t)+U(v,t))

[0050] Where Q(v,t) is the percentage of the remaining link capacity of the converter operating mode corresponding to the current node v in the previous time slot of the time slot t corresponding to its own level;

[0051] The newly selected optimal child node is used as the new parent node, and the number of times the converter working mode corresponding to all nodes in the search tree is selected as the optimal working mode is updated. Then, the confidence upper bounds of each child node of the selected new parent node are calculated, and the updated parent node is selected based on the confidence upper bounds. Similarly, the confidence upper bounds of the last-level leaf nodes are obtained, and the converter working mode corresponding to the leaf node with the largest confidence upper bound is used as the optimal working mode of the current converter in the next time slot.

[0052] Beneficial effects:

[0053] 1. The present invention provides a method for joint recognition and scheduling of heterogeneous information transmission. By integrating optical converters into the DCN topology, the network is endowed with the inherent ability to self-adjust according to traffic patterns. Compared with existing layered or extended graph solutions, the present invention has the ability to switch network topologies more flexibly, even during the ongoing data center service process.

[0054] 2. The present invention provides a method for joint recognition and scheduling transmission of non-homogeneous information, which cleverly integrates graph-based learning methods into DCN topology design and explores new possibilities for adapting to the dynamic characteristics of data center networks. This learning method can recognize non-homogeneous network information and make real-time scheduling decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A comparison diagram of the traditional CLOS topology and the converter-enhanced integration.

[0056] Figure 2 This is an example of an extended graph architecture and a PFC deadlock caused by CBD;

[0057] Figure 3 for network node information and its cognitive representation;

[0058] Figure 4 Schematic diagram of the overall framework of network cognition and switch scheduling decision process in MorpheusClos. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0060] To address the above challenges, we propose MorpheusClos, a name inspired by the ability of Morpheus in ancient Greek mythology to change and adapt. This innovative approach integrates converters into the existing CLOS architecture and uses fine-grained configuration strategies to improve DCN performance without requiring additional bandwidth resources. The core of MorpheusClos lies in its traffic pattern analysis method, which takes into account the topology of the network. Initially, we formulated the configuration problem as a Steiner tree problem. However, given the complexity of directly solving this NP-complete problem, we developed a learnable framework that brings the set of converters as close to the optimal solution as possible. The core of this framework is a temporal self-supervised method designed to handle the dynamic characteristics of DCNs. To the best of our knowledge, MorpheusClos is the first solution to explore the application of converters based on graph representation learning in DCN topology optimization.

[0061] The first contribution of this paper is the introduction of an innovative DCN topology that integrates switches with the existing CLOS architecture. This design leverages the switch's dynamic connection patterns to achieve diverse cross-pod connectivity and facilitate interconnection between different switch layers. By enhancing the adaptability of the CLOS network, it significantly improves the system's resilience to sudden traffic surges, enabling the physical topology to intuitively adapt to network traffic patterns.

[0062] The second contribution of this paper is the introduction of a graph representation learning framework specifically designed for DCNs. Compared to traditional graph representation learning methods, this framework demonstrates superior capabilities in handling dynamic, heterogeneous graph structures. By combining a temporally self-supervised Transformer with hierarchical aggregation of heterogeneous information, our model is able to effectively interpret and manage heterogeneous dynamic traffic information across multiple nodes.

[0063] Specifically, by integrating optical converters into the DCN topology, this invention empowers the network with the inherent ability to self-adjust based on traffic patterns. This section first explores the root causes of the performance differences between the traditional CLOS architecture and the extended graph when handling different traffic patterns. It then emphasizes the key role of the adaptive topology facilitated by optical converters in resolving these differences.

[0064] The hierarchical tree structure of the CLOS architecture provides redundancy and fault tolerance through multiple paths, providing strong fault tolerance. However, its hierarchical routing design results in more routing hops and higher transmission delays.

[0065] Extended graph architectures, such as Jellyfish, Xpander, and SlimFly, are being explored as alternatives to CLOS. With their flat topology, they offer advantages in bandwidth and latency. Figure 2 As shown in Figure 3, using a four-port switch, the extended graph reduces the average number of hops for communication between servers through flat random interconnection instead of a hierarchical structure, thereby improving the speed of data transmission between servers compared to CLOS.

[0066] To compare the performance of the CLOS and Extended Graph DCN architectures under different traffic scenarios, we created different traffic patterns using scripts in the iperf tool on the server. Our tests were conducted in a 40GbE data center with 100 switches and 1,400 servers, focusing on two main traffic patterns:

[0067] 1. All to All: Each server communicates with all other servers, simulating high load and intensive data exchange to test network capacity and fault tolerance.

[0068] 2. Uniform Random: The server randomly selects communication targets, similar to typical DCN traffic, allowing for realistic performance evaluation.

[0069] We summarized the performance of CLOS and Extended Graph topologies under various traffic and load conditions, highlighting their respective strengths. CLOS excels in all-to-all scenarios due to its redundant paths, making it suitable for high-load situations. On the other hand, the Extended Graph topology exhibits lower latency under simple traffic scenarios, but this advantage diminishes under higher loads, where CLOS performs better in terms of throughput. This demonstrates that while Extended Graph is better suited for typical traffic, CLOS excels under high-traffic demands. These results lead to a key conclusion: no single DCN topology is universally superior. The choice of topology should be closely tied to the specific traffic patterns and load conditions expected in the network environment.

[0070] Case Analysis of Deadlock Caused by Existing PFC Technology

[0071] Given the performance differences between different topologies, especially the degradation of scaling graph performance under high load, it is crucial to examine the impact of DCN's upper layer protocols (flow control algorithms at the transport layer and routing algorithms at the network layer).

[0072] Data center operators such as Microsoft and Alibaba are increasingly adopting Remote Direct Memory Access (RDMA) to improve network performance. A key component of RDMA is the Priority Flow Control (PFC) algorithm, a flow control method designed to prevent packet loss in high-speed data networks. PFC alleviates congestion by pausing specific data flows, ensuring that critical packets are not dropped.

[0073] Figure 2An example of PFC deadlock introduced in the extended graph architecture is shown. Ring buffer dependency (CBD) is a loop formed by the connections between value devices, which is avoided in tree topologies but is widely present in flat extended graph network frameworks. Figure 2 The following figure shows a CBD example in a network framework consisting of switches A, B, and C (connected by yellow lines). Although routing algorithms such as ECMP and KSP can prevent a single flow from forming a loop at the routing level, loop deadlocks can still occur in some special cases. For example, in Figure 2 In the example, data flow 1 is transmitted along the path C→B→A (red dashed line), while another flow travels along the path A→C (blue dashed line), forming a three-switch CBD loop. When the real-time data inventory in the buffer reaches the PFC threshold, this CBD loop becomes severe and can potentially cause a deadlock. A deadlock occurs when all nodes in the CBD loop simultaneously trigger the PFC threshold and suspend data forwarding. A deadlocked switch cannot recover, becoming an inactive node and reducing the overall transmission efficiency of the DCN. In contrast, the CLOS architecture, with its tree-like structure and hierarchical "up and down" routing architecture, avoids CBD loops, fundamentally preventing PFC deadlock and making it more adept at handling high-load traffic.

[0074] The above analysis shows that when running common upper-layer protocols, the extended graph architecture may experience deadlock within its CBD loop structure under high-load traffic patterns. However, the occurrence of CBD in an extended graph is not guaranteed; it depends on the prevailing traffic pattern. Furthermore, the presence of CBD does not necessarily mean that PFC-induced deadlock will be triggered. In less extreme traffic conditions, such as uniform random traffic patterns, CBD is uncommon. Even in the presence of CBD, deadlock will not occur if not all device caches within the CBD reach the PFC threshold. Conversely, in the absence of deadlock, the extended graph exhibits performance advantages over the CLOS architecture. Notably, our validation experiments employed the commonly used ECMP protocol, and similar trends were observed when testing other routing algorithms, including KSP and VLB. These observations lead us to conclude that when using switches to modify network topologies to resemble extended graph connections, traffic patterns and the number of flattened switches must be considered to minimize the risk of PFC-induced deadlock.

[0075] Furthermore, this paper will first briefly review previous works on heterogeneous DCNs with switches.

[0076] Two Paradigms for DCN Topology Design. Within the DCN topology landscape, two distinct approaches have emerged, each addressing specific network requirements and challenges. The first approach is the traditional static network topology, represented by structures such as Fat-Tree, VL2, Diamond, and BCube. These architectures primarily focus on scalability. For example, Fat-Tree is a tree-like topology that optimizes even traffic distribution and high bidirectional bandwidth. VL2 creates a logically complete bipartite graph by interconnecting intermediate and aggregation switches, simplifying cabling compared to Fat-Tree. Diamond enhances Fat-Tree by reducing average path length, while BCube provides a low-diameter, full-bandwidth architecture with multiple alternative paths.

[0077] Another trend is towards flat network architectures, such as Jellyfish, Xpander, and SlimFly. These designs represent a departure from traditional hierarchical structures. These topologies, based on extended graph theory, aim to create more flexible and cost-effective network designs. For example, Jellyfish employs a random graph topology, achieving high levels of connectivity and flexibility. Xpander builds on this foundation to provide a scalable network design. SlimFly proposes a low-diameter, high-bandwidth network. This DCN topology design direction is characterized by the pursuit of enhanced network efficiency.

[0078] Dynamic Network Topologies. Modern DCNs increasingly seek to increase flexibility by introducing randomness into their static topologies. Key research, such as Jellyfish, addresses the challenge of incremental scaling by injecting randomness into network design. Switches, emerging as a new class of devices with the advent of optical communications, bring dynamism to network architectures through a variety of configurations. Notable research, such as Flattree, demonstrates how switchable switches can transform traditional topologies into more random forms, thereby improving network performance. Dynamic architectures such as RotorNet and Opera embody this progress, featuring leaf-spine topologies with uplinks that are continuously reconfigured at the top-of-rack switch. However, these "dynamic" approaches are limited by their reliance on randomness and lack the ability to generate insights, analyze, and adapt to changing traffic patterns.

[0079] Graph-based learning methods. The development of graph embedding methods has opened up significant opportunities for graph analysis. These methods facilitate the understanding of complex graph structures by simplifying their representation. However, real-world networks often consist of multimodal and multi-type nodes and relationships, posing challenges in capturing their diverse characteristics. Recently, the focus in the field of graph representation learning has shifted to the design of graph transformers. These models, leveraging fully connected global attention mechanisms, have demonstrated remarkable results in various graph representation learning tasks. However, most existing graph transformers primarily focus on clustering and prediction based on existing data, making them less effective in handling dynamic, constantly evolving real-world networks such as DCNs. This poses a challenge in adapting these powerful models to complex and ever-changing networks. In this evolving field, MorpheusClos attempts to cleverly integrate graph-based learning methods into DCN topology design, exploring new possibilities for adapting to the dynamic nature of data center networks.

[0080] Specifically, a method for joint recognition and scheduling transmission of non-homogeneous information is provided, in which traffic is transmitted from a sending server to a receiving server via a meta-path composed of switches located at different layers of a heterogeneous data center network. The switches on the heterogeneous data center network are divided into core layer switches, aggregation layer switches, and edge layer switches. Adjacent switches on the meta-path are connected by converters, and the number of converters is the smaller of the number of aggregation layer switches and edge layer switches. At the same time, different operating modes of the converters correspond to different switch connection methods, and the operating mode of the converters changes with the traffic transmission requirements of the heterogeneous data center network. When it is necessary to change the traffic transmission meta-path of the heterogeneous data center network in the next time slot, the method for obtaining the operating mode of each converter in the next time slot is as follows:

[0081] S1: Obtaining status information of a heterogeneous data center network in a current time slot, wherein the heterogeneous data center network is composed of multiple converters, multiple switches, and multiple servers, and the status information includes service control information and topological location information of each converter in a current time slot, service control information and topological location information of each server in a current time slot, and topological location information and dynamic operation information of each switch in a current time slot;

[0082] It should be noted that if Figure 3As shown in Figure 3, the category of topological location information includes the connection and location information of DCN nodes, such as the core layer (C), aggregation layer (A), edge layer (E), and server (S). This information is represented in the form of an adjacency matrix and node attributes, capturing the complex relationships between nodes and dynamically reflecting topological updates within the MorpheusClos network. Dynamic operational information: This is defined as time-sensitive information about switching nodes in the DCN that cannot be determined in advance. This information is critical for real-time operational decisions and adaptive responses to network conditions. 3. Service and control information: This includes pre-accessible information related to servers, such as protocol type, service type, and function type, as well as control information of switches. These attributes play a guiding role in the intelligence of the network.

[0083] Strategies that exploit categorical information are crucial in addressing the heterogeneity of network information. Notably, traditional graph-based learning methods excel at deriving embeddings from static node features but often struggle to extract insights from dynamic information. Therefore, our approach is specifically designed to manage and exploit the dynamic behavior of DCNs within the MorpheusClos framework, emphasizing adaptability and real-time responsiveness.

[0084] A metapath represents a series of nodes that describes data transmission behavior. It includes the sending server, the receiving server, and a series of intermediate nodes determined by the routing algorithm. In traditional graph learning, information propagation is typically performed on a single hop basis. The concept of a metapath provides a higher-level perspective, encompassing multiple hops within a single, unified path. This higher-level abstraction facilitates easier understanding of network traffic events because it encapsulates the entire data transmission process from sender to receiver.

[0085] S2: Input the state information into the trained decision system to obtain the optimal working mode of each converter in the next time slot.

[0086] It should be noted that the Heterogeneous Data Center Network (HDCN) is modeled as a dynamic undirected graph Where V represents the switch, converter (that is, Figure 1 (b) The red line in the middle) and the diverse network nodes including the server. The set of connected edges between these elements is denoted as E. The adjacency matrix It can be dynamically allocated according to the real-time network conditions. At any time step t, the state of the network is determined by the graph signal , where f represents the number of node-specific features.

[0087] In MorpheusClos, the converter dynamically adjusts the network topology without requiring additional resources in the traditional CLOS architecture. A Aggregation layer switches and NumE edge layer switches, we define Num p =min(Num A ,Num E ). It represents the aggregation by placing a switch between p pairs of switches. Figure 4 As shown in the upper left corner, the default mode of a single switch in an AE switch pair is to replace the connection to the core layer switch and server with the switch. The remaining two ports of each switch are randomly connected to other switches in the network. The internal connections of the switch are reconfigurable, enabling cross-layer connections, such as between edge and core switches or between switches in different Pods. For a MorpheusClos network with p switches, the state of each switch can be represented as a two-dimensional matrix M:

[0088]

[0089] Each entry M i,j represents the state of the i-th converter in its j-th mode. The default mode of each converter is denoted as M i,1 ,When all converters are in default mode, the DCN topology is equivalent to a hierarchical CLOS ,architecture.

[0090] Furthermore, in step S2, the different operating modes of the switches, servers, and converters in the heterogeneous data center network are used as nodes of the heterogeneous data center network topology; at the same time, the decision system includes multiple cascaded Dyta modules, multi-layer perceptrons (MLPs), and PUCB models, wherein each Dyta module consists of a Dyta layer and a causal convolutional layer;

[0091] Except for the Dyta layer of the first-level Dyta module, the Dyta layers of the remaining Dyta modules are used to process the first node feature information stored in the form of a heterogeneous data center network topology structure obtained by the previous-level Dyta module based on the HDCN mutual attention mechanism to obtain second node feature information; the causal convolution layer is used to fuse the second node feature information obtained by the Dyta layer of the same Dyta module with the second node characteristic information of different time slots in history to obtain first node feature information used to describe the causal relationship between topology scheduling and network performance; wherein, the Dyta layer of the first-level Dyta module is used to convert state information into second node feature information based on the HDCN mutual attention mechanism, and the first node feature information obtained by the causal convolution layer of the last-level Dyta module is output to the multi-layer perceptron MLP;

[0092] The multilayer perceptron MLP is used to obtain the prior probability that each operating mode of each converter in the next time slot is the optimal operating mode based on the received first node feature information;

[0093] The PUCB model is used to calculate the corresponding upper confidence bounds of each converter when it is in various working modes in the next time slot based on the corresponding prior probabilities of each converter when it is in various working modes in the next time slot and the percentage of the remaining link capacity of each converter in the previous time slot, and the working mode corresponding to the maximum value of the lower confidence bound of each converter is used as the working mode of each converter in the next time slot.

[0094] Furthermore, the second node feature information obtained by the Dyta layer of any level Dyta module is composed of the target node output features corresponding to each converter, and the target node output features corresponding to any network node are obtained as follows:

[0095] Take each network node as the target node and perform the following steps to obtain the updated target node output features corresponding to each network node:

[0096] S21: Get the target node v t The adjacent nodes and the network nodes with similar traffic patterns in the previous time slot are used as source nodes v s ;

[0097] S22: Each source node v s The service control information and topological location information of the previous time slot are used as the original features of the source node, and the original features of the source node are reduced to obtain the source node input feature H (l-1) [v s ], where l represents the source node v s The hierarchical number of the Dyta module where it is located; the source node input feature of the first-level Dyta module is defined as H 0 [v s ];

[0098] S23: Set the target node v t The service control information and topological location information of the previous time slot are used as the original features of the target node, and the original features of each target node are reduced to obtain the target node input feature H (l-1) [v t ]; the target node input feature of the first-level Dyta module is defined as H 0 [v t ];

[0099] S24: Input each target node into the feature H through the query vector linear projection function Q-linear (l-1) [v t ] are mapped to multiple vector spaces respectively, and the query vector mapping Q of each vector space is obtainedh (v t ):

[0100] Q h (v t )=Q-linear h (H (l-1) [v t ])

[0101] Wherein, the vector space serial number h=1,2,...,H, H represents the number of vector spaces;

[0102] S25: Input the source node into the feature H through the key vector linear projection function K-linear (l-1) [v s ] is mapped to multiple vector spaces to obtain the key vector mapping K of each vector space h (v s ):

[0103] K h (v s )=K-linear h (H (l-1) [v s ])

[0104] S26: Input the source node into the feature H through the value vector linear projection function V-linear l [v s ] is mapped to multiple vector spaces to obtain the value vector mapping V of each vector space h (v s ):

[0105] V h (v s )=V-linear h (H (l-1) [v s ])

[0106] S26: Map K according to the key vector h (v s ) and query vector mapping Q h (v t ) Get the source node v s For each target node v t The importance of Attention(v s ,v t ):

[0107]

[0108] Among them, ATT_head h (v s ,vt ) represents any target node v in the hth vector space t and source node v s The attention weight between is calculated as:

[0109]

[0110] Where d represents the target node input feature H (l-1) [v t ] dimension, Indicates the use to capture the source node v s and target node v t The transformation matrix of the characteristic relationship between Represents the source node v s and target node v t The attention adaptive scaling factor between , T represents transposition;

[0111] S27: Using linear projection Map the value vector V h (v s ) is converted to a value message ValueMessage(v s ,v t ):

[0112]

[0113] Among them, MSG-head h (v s ,v t ) indicates that the set transformation matrix is ​​used Mapping V with value vector h (v s ) multiplied by, Indicates that all vector spaces correspond to the MSG-head h (v s ,v t ) added;

[0114] S28: The importance of Attention (v s ,v t ) and value message ValueMessage(v s ,v t ) to obtain each target node v t Message Vector

[0115]

[0116] Where SN(t) represents the set of source nodes selected in the current time slot t, It represents the product Attention(v) corresponding to the node pairs consisting of all source nodes and their paired target nodes in the set SN(t). s ,v t )·ValueMessage(v s ,v t ) added;

[0117] S29: Get the target node v according to the nonlinear activation ReLu function t The target node output feature H is output at the current level Dyta module (l) [t]:

[0118]

[0119] Among them, σ is the weight of the nonlinear activation ReLu function, H (l-1) [t] is the target node v t Output features at the target node output by the previous level Dyta module.

[0120] Furthermore, the method for the causal convolution layer of any level Dyta module to obtain the feature information of the first node is as follows:

[0121]

[0122] in, is the output feature of the rth causal convolution of the current causal convolution layer at the current time slot t, is the first node feature information finally obtained by the current causal convolution layer at the current time slot t, D is the expansion factor used to characterize the number of layers in the causal convolution layer, K is the number of convolutions performed in each layer of causal convolution, is the weight of the rth causal convolution layer in the current lth level Dyta module when performing the i-th convolution, G t-i×r represents the graph data within the expansion time step, where the graph data is obtained by treating the different working modes of each switch, server, and converter as a node to form a topology graph; the state information of the K×D topology graphs of the historical K×D time slots including the current time slot t is used as the second node feature information obtained in the current l-th level Dyta module as the graph data; σ represents the activation function, and ReLU represents the nonlinear activation ReLu function.

[0123] Furthermore, the method for determining the optimal operating mode of each converter in the next time slot is as follows:

[0124] Construct a multi-level search tree for the PUCB model, wherein the last-level leaf node of the search tree corresponds to the next time slot, and each of the remaining nodes corresponds to a historical time slot. Each node represents a converter operating mode and its related statistical information, wherein the statistical information includes the number of times it was selected as the optimal operating mode, the percentage of remaining link capacity in the previous time slot, and the prior probability. At the same time, except for the root node and leaf nodes, the remaining nodes serve as both parent nodes and child nodes.

[0125] The search tree is traversed according to the set search rules, and the confidence upper bounds of the converter operating modes corresponding to the nodes at each level are calculated level by level. The set search rules are that the nodes with fewer selections are selected first, and the nodes with larger confidence upper bounds are selected first. At the same time, the confidence upper bound of the converter operating mode corresponding to any node is obtained as follows:

[0126]

[0127] Where U(v,t) represents the confidence upper bound of the converter operating mode corresponding to the current node v in the time slot t corresponding to its own level, N(v,t) is the number of times the converter operating mode corresponding to the current node v is selected as the optimal operating mode before the end of the time slot t corresponding to its own level, P(v,t) is the highest a priori probability of the converter operating mode corresponding to the current node v in the time slot t corresponding to its own level, c puct is a setting constant that balances exploration and exploitation;

[0128] For each parent node, select the child node with the largest Q(v,t)+U(v,t) from its respective child nodes as the optimal child node:

[0129] Optimal child node ← argmax v (Q(v,t)+U(v,t))

[0130] Where Q(v,t) is the percentage of the remaining link capacity of the converter operating mode corresponding to the current node v in the previous time slot of the time slot t corresponding to its own level;

[0131] The newly selected optimal child node is used as the new parent node, and the number of times the converter working mode corresponding to all nodes in the search tree is selected as the optimal working mode is updated. Then, the confidence upper bounds of each child node of the selected new parent node are calculated, and the updated parent node is selected based on the confidence upper bounds. Similarly, the confidence upper bounds of the last-level leaf nodes are obtained, and the converter working mode corresponding to the leaf node with the largest confidence upper bound is used as the optimal working mode of the current converter in the next time slot.

[0132] The basic principle of the present invention is described in detail below.

[0133] 1. Problem Definition

[0134] The classic Steiner tree problem, which is considered NP-hard, is a basic combinatorial optimization problem. It is described as follows: Given a weighted graph G = (V, E) and a set of terminal nodes The goal is to find a minimum cost tree that contains all terminal nodes Te. Non-terminal nodes in the graph that do not belong to Te are called Steiner nodes.

[0135] In a specific scenario in the MorpheusClos framework, a switch in the network, represented by V Te , are considered as terminal nodes, while the various modes of the converter are represented as V C =VV Te , are regarded as potential Steiner nodes. The goal is to optimize the network configuration to minimize the maximum link utilization, which is a key metric for load balancing in a dynamic and heterogeneous DCN environment.

[0136] 2. MorpheusClos framework definition

[0137] In this section, we detail our network optimization strategy within the MorpheusClos framework, focusing on the role of converters. Our approach is based on a comprehensive analysis of DCN topologies and exploits the rich graph structure of these networks. The core advantages of this exploration include: (1) Graph-driven insights: We investigate the ability of graph learning methods to solve Steiner tree-like problems, leveraging the structural complexity of DCNs. (2) Improved accuracy: By leveraging the inherent heterogeneous information in DCN nodes and their connections, our approach surpasses traditional decision-making processes based on raw data. (3) Adaptive online decision-making: Our strategy dynamically represents node relationships using a weighted adjacency matrix, enabling responsive and evolving decision-making under real-time network conditions.

[0138] Figure 4 (a) shows the overall process of converter mode decision in the MorpheusClos framework.

[0139] First, we generate a preliminary adjacency matrix to represent the topological location information of the graph. Node attributes and most service and control information are one-hot encoded to form the initial attributes. Due to the time-sensitive nature of dynamic operation information, it is not included in the node attributes. Instead, it is used to calculate node similarity scores and integrated with the adjacency matrix as weights. Formally, we represent the HDCN data sequence for time period tT to t as follows:

[0140]

[0141] The adjacency matrix and graph information of these T time periods are input into the network composed of DyTa (Dynamic Traffic Attention) blocks. Subsequently, the embeddings processed by the DyTa layer and the causal convolution layer are concatenated and aggregated along the time dimension to form a comprehensive representation of the DCN node information and traffic patterns at time t:

[0142]

[0143] The next step is to Pass through a multi-layer perceptron (MLP) layer to derive the effective prior probability of each converter mode node:

[0144]

[0145] These probabilities are converted into the converter configuration for the next time period t+1 using a probabilistic update confidence bound (PUCB) model. We employ an iterative update scheme for adaptive decision making, treating the HDCN atlas as a FIFO stack with depth T and updating it after topology adjustments. This allows converter decisions to be executed cyclically during the framework's execution.

[0146] 2.1. Enhanced Graph Analysis with Real-Time Network Data

[0147] Dynamic weighting of the adjacency matrix: In a dynamic network environment, nodes continuously generate operational information with significant time sensitivity. Traditional graph-based learning models have difficulty in effectively processing this information. To ensure that traffic patterns and other operational data effectively guide decision-making, we incorporate these elements into the adjacency matrix of HDCN. For each terminal node, we consider its operational information attribute sequence (o t-T,n ,...,o t,n ), from (O t-T ,...,O t ) creates an overall representation of the node's dynamic operation information attributes:

[0148]

[0149] u n Represents the aggregated operation information of the nth terminal node (the switch is the terminal), based on the derived aggregation weight p τ,n Here, τ indexes the time period (tT,t). The aggregation weight p τ,n Reflects the traffic pattern of a specific node (o τ,n ) and the overall traffic pattern (u n ) between the two vectors. is a learnable parameter vector for the transformation, where D represents the dimension of the terminal node attributes.

[0150] We propose to estimate the degree of heterogeneity between nodes, which reflects the variability of their traffic distribution over time. This can be mathematically expressed as follows:

[0151]

[0152] Here, the higher q m,n Indicates that the traffic patterns between nodes m and n are more similar. This metric is integrated as a weight in the adjacency matrix of graph G to form a weighted adjacency matrix

[0153] 3. Joint Cognition via Heterogeneous Graph Transformers

[0154] We exploit the structural properties of the graph to compute mutual attention between nodes and facilitate information transfer. This approach makes it possible to learn embeddings of transformer nodes in the context of the network.

[0155] 3.1 DyTa layer

[0156] like Figure 4 As shown in (b), the DyTa layer is a key component of the MorpheusClos framework, and its unique ability lies in generating attention masks for heterogeneous information during message passing.

[0157] Adaptive Node Selection: In DCN data, traffic often passes through multiple nodes, resulting in long meta-paths. Capturing these extended relationships effectively is a challenge for graph-based learning tools. To address this issue, we introduce a node selection module before the message passing process. This module selects the source node for message passing based on two criteria: physical connectivity with the current node and presence of a node in the weighted adjacency matrix. The mathematical formula of the node selection process is as follows:

[0158]

[0159] The selection criteria for message passing nodes are based on two key aspects. First, N(i) represents the set of physical neighbor nodes of node i, ensuring that local structural connections play a role in the message propagation process. Second, the inclusion of node k is determined by its weighted adjacency matrix The high operational similarity with node i is determined in this matrix. and denote the mean and standard deviation of the similarity scores for row i, respectively. Nodes with similarity scores exceeding two standard deviations above the mean are considered significantly similar to node i. This threshold is chosen to capture nodes that are relevant to the operation, ensuring that only nodes with significant similarity in traffic patterns to node i are included.

[0160] HDCN mutual attention mechanism: The DyTa layer aims to update the embedding of the target node. Figure 4 As shown in (b), the embedding of the target node is converted into a query vector (Q) through a linear layer, while the embedding of the selected node as the source node is converted into a key vector (K) and a value vector (V). This mechanism differs from traditional transformers because it takes into account the heterogeneity of node information. Each pair of node relationships has its own unique projection weight, which is shared between pairs of nodes of the same type.

[0161] The embedding of network nodes in different DyTa layers is represented as H l , where l represents the lth DyTa layer. In the DyTa layer, the embedding of a node is affected by its selected node set SN. To describe this relationship, we refer to the nodes in the selected node set as source nodes v s , the affected node is called the target node v t . Target node v t and the embedding of the source node at layer l are denoted as H l [v t ] and H l [v s In order to efficiently manage the information flow within the network, the DyTa layer adopts a multi-head attention mechanism. t and v s The embedding of is divided into h heads, allowing the model to focus on information in different embedding subspaces in parallel. This parallel processing not only reduces computation time but also allows for a more nuanced understanding of the network dynamics.

[0162] These node embeddings are mapped into the h-th vector space using three linear projection functions:

[0163] Q h (v t )=Q-linear h (H (l-1) [v t ])

[0164] K h (v s )=K-linear h (H (l-1) [v s ])

[0165] V h (v s )=V-linear h (H (l-1) [v s ])

[0166] To calculate the node v t and v smutual attention between them, we introduce an attention operator to estimate the s v t Importance:

[0167]

[0168] The attention function can be described as mapping the query vector and a set of key-value pairs to each pair of nodes (v s ,v t ) output. t and v s The overall attention is the connection of the attention weights of all heads, processed by the softmax function. h (v s ,v t ) represents the h-th head v t and v s The attention weight between is calculated as follows:

[0169]

[0170] here, is a transformation matrix used to capture feature relationships, and d is the output H of the previous DyTa layer (l-1) [v t ] dimension, Serves as an adaptive scaling factor for attention. The final step involves concatenating the h attention heads to form an attention vector for each pair of nodes. Subsequently, for the target node t, the attention vectors of all selected nodes SN(t) are compiled and normalized via a softmax operation. This ensures that the sum of the attention weights of all nodes in SN(t) equals one, adhering to the principle of distributed attention.

[0171] Message passing in the DyTa layer: Figure 4 As shown in (b), when calculating all node pairs (v s ,v t ), these weights are combined with the value vector (V) to update the target node v t The value message for each source-target node pair is defined as follows:

[0172]

[0173] To construct the i-th message header MSG-headi(s,t), the features of the source node s are projected to the i-th message vector using linear projection Then through the matrix Adjust the original V h (v s) feature vector. The last step is to aggregate all h message headers to obtain the value vector of each pair of nodes (s, t). This method uses a method similar to that of generating attention weights, introducing the transformation matrix To balance the differences in feature distribution of heterogeneous nodes.

[0174] After computing the heterogeneous multi-head attention and value messages, we need to aggregate them from the source node to the target node (see Figure 4 (b)). Since the softmax process in the formula normalizes the sum of the attention vectors to one, we can simply use the attention vectors as weights to average the corresponding messages from the source node to update the vector as follows:

[0175]

[0176] This aggregates information from the selected set of nodes SN(t) to the target node v t We update the vector Apply the nonlinear activation ReLu function as follows:

[0177]

[0178] In this way, we get the output H of the target node t in the DyTa layer (l) [t]. By stacking modules containing DyTa layers, the information of each node is allowed to reach most nodes in the HDCN graph. Therefore, the DyTa layer generates a representation of each node that contains global traffic pattern features for downstream tasks of predicting converter priors.

[0179] Throughout the DyTa layer architecture, we rely heavily on the weighted adjacency matrix Generate attention weights. This can be interpreted as leveraging traffic patterns to jointly update the embeddings of multiple nodes to the target node. Compared to standard graph transformers, our DyTa layer distinguishes the relationship weights between heterogeneous nodes, and thus better handles distribution differences in HDCN.

[0180] 3.2 Temporal Aggregation via Causal Convolution

[0181] As part of the DyTa block, the task of the causal convolution layer is to process the temporal HDCN dataset as shown in formula (1) to ensure that the causal order of the provided data is preserved. For one-dimensional temporal data, this operation can be achieved by ordinary convolution, but to achieve a sufficiently large receptive field, multiple layers of convolution need to be stacked. In order to reduce the computational cost, dilated convolution is used. Using a certain dilation step size d, the advantage of dilated convolution is that it can exponentially expand the receptive field while maintaining high computational efficiency. The process of dilated causal convolution can be defined as:

[0182]

[0183] Represents the output features of time t and layer r. This generalizes the dilated causal convolution operation, where K is the convolution kernel size and D is the dilation factor. is the convolution weight of layer l, G t-i×r represents the graph data at the dilated time steps. ReLU introduces nonlinearity in the convolution process. Finally, σ(·) is the nonlinear activation function applied to the convolution output, ensuring that the model captures the temporal dynamics of the graph data while preserving causal relationships.

[0184] Compared to recurrent neural networks (RNNs), dilated causal convolution allows for parallel computation and, due to its non-recursive nature, alleviates the problem of gradient vanishing or exploding. Considering its superiority in handling time series problems, we use this module to process the time evolution information of the graph. Figure 4 As shown in (c), past graph snapshots are aggregated into a single heterogeneous data center information topology set, which encapsulates information from the past T time periods, enabling the MorpheusClos framework to effectively process information from multiple past time periods and dynamically understand traffic trends in DCNs.

[0185] 4. Graph-enhanced PUCB for converter mode decision

[0186] In the previous section, we used graph transformers to obtain prior probabilities for identifying the best transformer patterns (Steiner nodes). A naive approach is to rank the patterns of each transformer based on these prior probabilities. However, this approach can be unreliable because the learning-based model only has one chance to calculate the best solution per time period and does not backtrack or reverse decisions. Furthermore, the decision space grows exponentially as the number of transformers increases, and decisions cannot be made in parallel because the decisions of different transformers influence each other.

[0187] This section outlines the decision-making process for outputting prior probabilities that guide converter mode decisions in the MorpheusClos framework. We introduce the exploration and exploitation balance principle of proportional confidence upper bound (PUCB) in this step. We construct a search tree containing converter mode nodes embedded with prior probabilities, ensuring that different mode nodes of the same converter cannot be visited simultaneously. Each node of the search tree in time period t is represented as a triple (N(v,t),Q(v,t),P(v,t)), where N(v,t) is the number of visits, Q(v,t) is the percentage of the remaining link capacity of the current node, and P(v,t) is the highest prior probability of the node.

[0188] PUCB decision process: In the context of the Steiner tree problem, our goal is to explore configurations that minimize link utilization. The decision process consists of several stages: Selecting a strategy: The process starts from the root node, and each step is based on a strategy that balances exploration (visiting the converter pattern suggested by the prior probability) and utilization (visiting the converter pattern of the previous time period). We define the overall reward of a node as a combination of its prior probability and the percentage of remaining link capacity. The method for determining the child node (actually referring to the converter pattern, not the server, switch, converter mentioned above) is as follows:

[0189] children node←argmax v (Q(v,t)+U(v,t))

[0190]

[0191] Among them, c puct is a constant that balances exploration and exploitation. Here, U(v,t) represents the confidence bound for node v. Recursive Decision Making: After a child node is selected, this decision translates into a change in the DCN topology. The remaining nodes in the search tree are then reevaluated. The selected child node becomes the root node for the next phase of the search process, which involves recursively computing the confidence bounds for its child nodes and selecting the next node based on these calculations. State Update: Once all switches have made a decision, the statistics for each node in the search tree are updated, including the number of visits N(v), the percentage of remaining link capacity Q(v), and the prior probability P(v).

[0192] Scheduling and Adaptation: Our framework optimizes the switch configuration to evenly distribute network traffic. Figure 4 (d) shows how changes in the confidence bounds of the switch nodes lead to topological adjustments, such as the red switch on the right changing its connection pattern in response to a decrease in the confidence bounds of its current mode nodes. This approach is particularly effective in dynamic and heterogeneous distributed communication networks.

[0193] 5. Practical considerations

[0194] The workflow consists of the following stages. First, the server proactively sends transmission requests and service information to the switch and controller. Subsequently, the switch and converter collaborate on routing, with the switch sending dynamic operation information to the converter. Third, the controller collects information from the converter to update the heterogeneous graph information set. Finally, the controller propagates the converter's pattern decision information, and routing in the switch is reconfigured in response to the topology change.

[0195] Decision Cycle: First, considering the stability of data center workloads, we chose to perform sensing and decision-making periodically. Based on experience, we set each cycle to 90 seconds, consisting of a control phase and a longer uptime period, approximately 50 times the control phase, to achieve a balance between overhead and adaptability. During the control phase, the switch's optical link is temporarily disconnected, and traffic needs to be buffered and wait for the next allocation.

[0196] Time cost of topology changes: Time cost is another key factor affecting the overall performance of DCNs. This includes the time cost of adjusting OpenFlow rules and the time cost of physical switch reconfiguration. Experiments have shown that the core bandwidth of the network decreases after a topology change and quickly recovers to a stable state within 1.5 to 2.5 seconds. Considering the network bandwidth benefits of topology changes, the time cost impact on bandwidth is approximately 2%, which is acceptable.

[0197] In summary, the first innovation of this invention is the introduction of a novel DCN topology that integrates switches with the existing CLOS architecture. This design leverages the switch's dynamic connection mode to achieve diverse cross-pod connectivity and facilitate interconnection between different switch layers. By enhancing the adaptability of the CLOS network, it significantly improves the system's resilience to sudden traffic surges, enabling the physical topology to intuitively adapt to network traffic patterns.

[0198] The second innovation of this paper is the introduction of a graph representation learning framework specifically designed for heterogeneous networks. Compared to traditional graph representation learning methods, this framework demonstrates superior capabilities in handling dynamic heterogeneous graph structures. By combining a temporally self-supervised Transformer with the hierarchical aggregation of heterogeneous information, our model is able to effectively interpret and manage heterogeneous dynamic traffic information across multiple nodes.

[0199] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may of course make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for joint recognition and scheduling transmission of heterogeneous information, characterized in that: Traffic is transmitted from the sending server to the receiving server via a meta-path composed of switches located at different layers of the heterogeneous data center network. The switches on the heterogeneous data center network are divided into core layer switches, aggregation layer switches, and edge layer switches. Adjacent switches on the meta-path are connected by converters, and the number of converters is the smaller of the number of aggregation layer switches and edge layer switches. Different operating modes of the converters correspond to different switch connection methods, and the operating mode of the converters changes with the traffic transmission requirements of the heterogeneous data center network. Therefore, when it is necessary to change the traffic transmission meta-path of the heterogeneous data center network in the next time slot, the method for obtaining the operating mode of each converter in the next time slot is as follows: S1: Obtaining status information of a heterogeneous data center network in a current time slot, wherein the heterogeneous data center network is composed of multiple converters, multiple switches, and multiple servers, and the status information includes service control information and topological location information of each converter in a current time slot, service control information and topological location information of each server in a current time slot, and topological location information and dynamic operation information of each switch in a current time slot; S2: Inputting the state information into the trained decision system to obtain the optimal operating mode of each converter in the next time slot; In step S2, the different operating modes of the switches, servers, and converters in the heterogeneous data center network are used as nodes in the heterogeneous data center network topology; at the same time, the decision system includes multiple cascaded Dyta modules, multi-layer perceptrons (MLPs), and PUCB models, wherein each Dyta module consists of a Dyta layer and a causal convolutional layer; Except for the Dyta layer of the first-level Dyta module, the Dyta layers of the remaining Dyta modules are used to process the first node feature information stored in the form of a heterogeneous data center network topology structure obtained by the previous-level Dyta module based on the HDCN mutual attention mechanism to obtain second node feature information; the causal convolution layer is used to fuse the second node feature information obtained by the Dyta layer of the same Dyta module with the second node characteristic information of different time slots in history to obtain first node feature information used to describe the causal relationship between topology scheduling and network performance; wherein, the Dyta layer of the first-level Dyta module is used to convert state information into second node feature information based on the HDCN mutual attention mechanism, and the first node feature information obtained by the causal convolution layer of the last-level Dyta module is output to the multi-layer perceptron MLP; The multilayer perceptron MLP is used to obtain the prior probability that each operating mode of each converter in the next time slot is the optimal operating mode based on the first node feature information output by the previous stage Dyta module; The PUCB model is used to calculate the corresponding upper confidence bounds of each converter when it is in various working modes in the next time slot based on the corresponding prior probabilities of each converter when it is in various working modes in the next time slot and the percentage of the remaining link capacity of each converter in the previous time slot, and to use the working mode with the maximum upper confidence bound value among multiple modes of each converter as the working mode of each converter in the next time slot.

2. The method for joint recognition and scheduling transmission of heterogeneous information according to claim 1, characterized in that: The second node feature information obtained by the Dyta layer of any level Dyta module is composed of the target node output features corresponding to each node in the network, and the method for obtaining the target node output features corresponding to any network node is: Take each network node as the target node and perform the following steps to obtain the updated target node output features corresponding to each network node: S21: Get the target node v t The adjacent nodes and the network nodes with similar traffic patterns in the previous time slot are used as source nodes v s ; S22: Each source node v s The service control information and topological location information of the previous time slot are used as the original features of the source node, and the original features of the source node are reduced to obtain the source node input feature H (l-1) [v s ], where l represents the source node v s The hierarchical number of the Dyta module where it is located; the source node input feature of the first-level Dyta module is defined as H 0 [v s ]; S23: Set each target node v t The service control information and topological location information of the previous time slot are used as the original features of the target node, and the original features of each target node are reduced to obtain the target node input feature H (l-1) [v t ]; the target node input feature of the first-level Dyta module is defined as H 0 [v t ]; S24: Input each target node into the feature H through the query vector linear projection function Q-linear (l-1) [v t ] are mapped to multiple vector spaces respectively, and the query vector mapping Q of each vector space is obtained h (v t ): Q h (v t )=Q-linear h (H (l-1) [v t ]) Wherein, the vector space serial number h=1,2,...,H, H represents the number of vector spaces; S25: Input the source node into the feature H through the key vector linear projection function K-linear (l-1) [v s ] is mapped to multiple vector spaces to obtain the key vector mapping K of each vector space h (v s ): K h (v s )=K-linear h (H (l-1) [v s ]) S26: Input the source node into the feature H through the value vector linear projection function V-linear l [v s ] is mapped to multiple vector spaces to obtain the value vector mapping V of each vector space h (v s ): V h (v s )=V-linear h (H (l-1) [v s ]) S26: Map K according to the key vector h (v s ) and query vector mapping Q h (v t ) Get the source node v s For each target node v t The importance of Attention(v s ,v t ): Among them, ATT_head h (v s ,v t ) represents any target node v in the hth vector space t and source node v s The attention weight between is calculated as: Where d represents the target node input feature H (l-1) [v t ] dimension, Indicates the use to capture the source node v s and target node v t The transformation matrix of the characteristic relationship between Represents the source node v s and target node v t The attention adaptive scaling factor between , T represents transposition; S27: Using linear projection M-Linear h : Map the value vector V h (v s ) is converted to a value message ValueMessage(v s ,v t ): Among them, MSG-head h (v s ,v t ) indicates that the set transformation matrix is ​​used Mapping V with value vector h (v s ) multiplied by, Indicates that all vector spaces correspond to the MSG-head h (v s ,v t ) added; S28: The importance of Attention (v s ,v t ) and value message ValueMessage(v s ,v t ) to obtain each target node v t Message Vector Where SN(t) represents the set of source nodes selected in the current time slot t, It represents the product Attention(v) corresponding to the node pairs consisting of all source nodes and their paired target nodes in the set SN(t). s ,v t )·ValueMessage(v s ,v t ) added; S29: Get the target node v according to the nonlinear activation ReLu function t The target node output feature H is output at the current level Dyta module (l) [t]: Among them, σ is the weight of the nonlinear activation ReLu function, H (l-1) [t] is the target node v t Output features at the target node output by the previous level Dyta module.

3. The method for joint recognition and scheduling transmission of heterogeneous information according to claim 1, characterized in that: The method for obtaining the first node feature information of the causal convolution layer of any level Dyta module is as follows: in, is the output feature of the rth causal convolution of the current causal convolution layer at the current time slot t, is the first node feature information finally obtained by the current causal convolution layer at the current time slot t, D is the expansion factor used to characterize the number of layers in the causal convolution layer, K is the number of convolutions performed in each layer of causal convolution, is the weight of the rth causal convolution layer in the current lth level Dyta module when performing the i-th convolution, G t-i×r represents the graph data within the expansion time step, where the graph data is obtained by treating the different working modes of each switch, server, and converter as a node to form a topology graph; the state information of the K×D topology graphs of the historical K×D time slots including the current time slot t is used as the second node feature information obtained in the current l-th level Dyta module as the graph data; σ represents the activation function, and ReLU represents the nonlinear activation ReLu function.

4. The method for joint recognition and scheduling transmission of heterogeneous information according to claim 1, characterized in that: The optimal operating mode of each converter in the next time slot is determined as follows: Construct a multi-level search tree for the PUCB model, wherein the last-level leaf node of the search tree corresponds to the next time slot, and each of the remaining nodes corresponds to a historical time slot. Each node represents a converter operating mode and its related statistical information, wherein the statistical information includes the number of times it was selected as the optimal operating mode, the percentage of remaining link capacity in the previous time slot, and the prior probability. At the same time, except for the root node and leaf nodes, the remaining nodes serve as both parent nodes and child nodes. The search tree is traversed according to the set search rules, and the confidence upper bounds of the converter operating modes corresponding to the nodes at each level are calculated level by level. The set search rules are that the nodes with fewer selections are selected first, and the nodes with larger confidence upper bounds are selected first. At the same time, the confidence upper bound of the converter operating mode corresponding to any node is obtained as follows: Where U(v,t) represents the confidence upper bound of the converter operating mode corresponding to the current node v in the time slot t corresponding to its own level, N(v,t) is the number of times the converter operating mode corresponding to the current node v is selected as the optimal operating mode before the end of the time slot t corresponding to its own level, P(v,t) is the highest a priori probability of the converter operating mode corresponding to the current node v in the time slot t corresponding to its own level, c puct is a setting constant that balances exploration and exploitation; For each parent node, select the child node with the largest Q(v,t)+U(v,t) from its respective child nodes as the optimal child node: Optimal child node ← argmax v (Q(v,t)+U(v,t)) Where Q(v,t) is the percentage of the remaining link capacity of the converter operating mode corresponding to the current node v in the previous time slot of the time slot t corresponding to its own level; The newly selected optimal child node is used as the new parent node, and the number of times the converter working mode corresponding to all nodes in the search tree is selected as the optimal working mode is updated. Then, the confidence upper bounds of each child node of the selected new parent node are calculated, and the updated parent node is selected based on the confidence upper bounds. Similarly, the confidence upper bounds of the last-level leaf nodes are obtained, and the converter working mode corresponding to the leaf node with the largest confidence upper bound is used as the optimal working mode of the current converter in the next time slot.

Images