Information processing and control method and device based on computing power scheduling and electronic equipment

Abstract

This disclosure presents an information processing and control method, apparatus, and electronic device based on computing power scheduling. One specific implementation of the method includes: in response to the existence of a computing task to be executed, acquiring real-time network status information; generating task scheduling information based on task description information corresponding to the computing task to be executed and the real-time network status information; generating first heat dissipation control information based on the task scheduling information, first node description information, and a pre-trained temperature prediction model; and performing heat dissipation pre-control on the network nodes involved in the task scheduling information based on the first heat dissipation control information. This implementation achieves heat dissipation control under efficient computing power scheduling.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to information processing and control methods, apparatuses, and electronic devices based on computing power scheduling. Background Technology

[0002] With the development of artificial intelligence technology, computing tasks are increasingly characterized by large data processing volumes and complex calculations. Currently, increasing the number of hardware components (e.g., increasing the number of GPUs (Graphics Processing Units) and NPUs (Neural Processing Units)) can improve the processing power of computing tasks. However, as the number of hardware components increases, communication bottlenecks become increasingly apparent. Simultaneously, the large amount of heat generated by the hardware can also affect its normal operation.

[0003] As computing workloads increase, computer clusters are evolving towards denser arrays, exacerbating heat generation issues. Failure to effectively control temperature can lead to problems such as frequency throttling, impacting task execution. Furthermore, prolonged exposure to high temperatures reduces hardware lifespan and increases the probability of failure. Currently, dedicated air conditioning systems in data centers can cool the entire space containing the hardware, but they are insufficient for effectively cooling localized areas (e.g., edge locations, areas inaccessible by direct airflow). While lowering temperature thresholds can reduce ambient temperature, this results in high power consumption for the air conditioning system. Therefore, this disclosure proposes a refined hardware (node) temperature control method incorporating computing power scheduling. Since the hardware is the carrier of computing tasks, this computing power scheduling approach enables refined temperature control, precise point-to-point temperature regulation, and effectively reduces power consumption during temperature control. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure propose information processing and control methods, apparatuses, and electronic devices based on computing power scheduling to solve the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide an information processing and control method based on computing power scheduling. The method includes: in response to the existence of a computing task to be executed, collecting real-time network status information, wherein the real-time network status information characterizes the real-time network status of a computing power scheduling network, and wherein the node types corresponding to network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type; generating task scheduling information based on task description information corresponding to the computing task to be executed and the real-time network status information, wherein the task scheduling information includes: task routing description information and task execution description information; generating first heat dissipation control information based on the task scheduling information, first node description information, and a pre-trained temperature prediction model, wherein the first node description information characterizes the node temperature state of the network nodes involved in the task scheduling information when the computing task to be executed has not yet started execution; and performing heat dissipation pre-control on the network nodes involved in the task scheduling information based on the first heat dissipation control information.

[0007] Secondly, some embodiments of this disclosure provide an information processing and control apparatus based on computing power scheduling. The apparatus includes: a collection unit configured to collect real-time network status information in response to the existence of a computing task to be executed, wherein the real-time network status information characterizes the real-time network status of a computing power scheduling network, and wherein the node types corresponding to the network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type; a first generation unit configured to generate task scheduling information based on task description information corresponding to the computing task to be executed and the real-time network status information, wherein the task scheduling information includes: task routing description information and task execution description information; a second generation unit configured to generate first heat dissipation control information based on the task scheduling information, first node description information, and a pre-trained temperature prediction model, wherein the first node description information characterizes the node temperature status of the network nodes involved in the task scheduling information when the computing task to be executed has not started execution; and a heat dissipation pre-control unit configured to perform heat dissipation pre-control on the network nodes involved in the task scheduling information based on the first heat dissipation control information.

[0008] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0009] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0010] The above embodiments of this disclosure have the following beneficial effects: Through the information processing and control method based on computing power scheduling in some embodiments of this disclosure, heat dissipation control under efficient computing power scheduling is achieved. Specifically, firstly, in response to the existence of a computing task to be executed, real-time network status information is collected. This real-time network status information represents the real-time network status of the computing power scheduling network. The node types corresponding to the network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type. This obtains the network communication status of the computing power scheduling network. Secondly, based on the task description information corresponding to the computing task to be executed and the real-time network status information, task scheduling information is generated. This task scheduling information includes: task routing description information and task execution description information. This generates task scheduling information matching the computing task to be executed from the perspective of task routing path and computing power allocation. Next, based on the task scheduling information, the first node description information, and a pre-trained temperature prediction model, first heat dissipation control information is generated. The first node description information represents the node temperature status of the network nodes involved in the task scheduling information when the computing task to be executed has not yet started execution. Finally, based on the aforementioned first heat dissipation control information, heat dissipation pre-control is performed on the network nodes involved in the aforementioned task scheduling information. In this way, heat dissipation pre-control of network nodes is achieved by combining task scheduling and node temperature status, thereby ensuring that network nodes execute computational tasks at optimal ambient temperatures, thus achieving the goal of heat dissipation control under efficient computing power scheduling. Attached Figure Description

[0011] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0012] Figure 1 This is a flowchart of some embodiments of the information processing and control method based on computing power scheduling according to the present disclosure;

[0013] Figure 2 This is a schematic diagram of the network structure of the computing power scheduling network;

[0014] Figure 3 This is a schematic diagram of the task feature generation process;

[0015] Figure 4This is a schematic diagram of the target routing path;

[0016] Figure 5 These are schematic diagrams of some embodiments of the information processing and control device based on computing power scheduling according to this disclosure;

[0017] Figure 6 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0018] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0019] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0020] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0021] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0023] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] refer to Figure 1 The diagram illustrates a flowchart 100 of some embodiments of the information processing and control method based on computing power scheduling according to the present disclosure. This information processing and control method based on computing power scheduling includes the following steps:

[0025] Step 101: In response to the existence of a computational task to be executed, collect real-time network status information.

[0026] In some embodiments, the executing entity (e.g., a computing device) of the information processing and control method based on computing power scheduling can collect real-time network status information in response to the existence of computing tasks to be executed. This real-time network status information characterizes the real-time network status of the computing power scheduling network. In practice, the real-time network status information can be composed of node status information corresponding to the network nodes included in the computing power scheduling network. The node status information can include: node identifier, node type, node latency, node bandwidth, node speed, and node communication address. Specifically, for network nodes whose node type is a task execution node, the corresponding node status information can also include: total computing resources and total remaining computing resources. The computing power scheduling network can be a communication network used for computing power scheduling. The computing power scheduling network can be composed of multiple network nodes. The node types corresponding to the network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type. In practice, network nodes of the scheduling node type can be used to receive computing tasks to be processed and to initiate computing power scheduling for those tasks. Network nodes of the routing node type can be used to forward computing tasks. Network nodes of type task execution node can be used to execute computational tasks.

[0027] In practice, network nodes of the scheduling node type can be used to monitor whether users initiate computing tasks. When a computing task to be executed is detected, real-time network status information is collected. Specifically, for example, when the concurrency of computing tasks is low, scheduling node-type network nodes can proactively obtain the node status information of other network nodes to obtain real-time network status information. Conversely, when the concurrency of computing tasks is high, a timed broadcast method can be used, where network nodes in the computing power scheduling network periodically broadcast their corresponding node status information. In this case, scheduling node-type network nodes obtain real-time network status information by capturing the broadcast node status information. Furthermore, monitoring tools (such as Zabbix) can be deployed in the computing power scheduling network, and a proxy method can be used to collect the node status information of the corresponding network nodes.

[0028] As an example, see Figure 2 The diagram shows the network structure of the computing power scheduling network, which includes one network node 201 of type scheduling node, two network nodes 202 of type task execution node, and eight network nodes 203 of type routing node.

[0029] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or terminal device. When the computing device is software, it can be installed within the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here. In particular, the computing device can be a network node included in a computing power scheduling network. For example, network nodes of the scheduling node type and network nodes of the task execution node type.

[0030] Step 102: Generate task scheduling information based on the task description information and real-time network status information corresponding to the computation task to be executed.

[0031] In some embodiments, the aforementioned execution entity can generate task scheduling information based on the task description information and real-time network status information corresponding to the computation task to be executed. The task scheduling information includes task routing description information and task execution description information. The task routing description information can represent the routing path from a network node of type scheduling node to a network node of type task execution node used to process the computation task to be executed. The task execution description information can represent the computing resources allocated to the computation task to be executed. The task description information can represent the computing power requirements of the computation task to be executed. For example, the task description information may include: task identifier, task data volume, maximum tolerable latency of the task, and bandwidth required by the task. In practice, a multimodal model can be used to extract features from the task description information and real-time network status information to generate task scheduling information specific to the computation task to be executed.

[0032] In some optional implementations of certain embodiments, the execution entity generates task scheduling information based on the task description information corresponding to the computational task to be executed and the real-time network status information, including:

[0033] Step S1: Extract task features from the above task description information using a task feature extraction model to generate task features.

[0034] In practice, the task feature extraction model includes M Transformer Blocks. For example, M can be 32. These M Transformer Blocks are connected serially. A random controller is configured for each Transformer Block from M / 2 to M-1. This random controller controls whether the output of the Transformer Block is included as part of the task feature extraction model's output. The random controller uses corresponding output probabilities for this control.

[0035] As an example, see Figure 3 The diagram shows the process of generating task features. First, the task description information is encoded using word encoding to obtain a set of word vectors 301. Figure 3 In this context, E1, E2…En represent word vectors in word vector set 301. Then, word vector set 301 is input to task feature extraction model 302. Task feature extraction model 302 includes M Transformer Blocks. Each Transformer Block with index M / 2 to M-1 is assigned a random controller 303. Specifically, as the index increases, the output probability of the random controller corresponding to the Transformer Block gradually increases. For example, the output probability of the Transformer Block with index M-1 is greater than the output probability of the Transformer Block with index M / 2. Then, the outputs of the Transformer Blocks with indices M / 2 to M are concatenated to obtain the initial task feature 304. The feature dimension of the initial task feature 304 is (MM / 2+1)×N. Since the Transformer Blocks with indices M / 2 to M-1 are controlled by output probability, when the output probability representation is not output, the corresponding feature of the Transformer Block in the initial task feature is 0. Next, considering the presence of zero features in the initial task features, feature compression is performed to reduce feature dimensionality, resulting in task feature 305. The reason for not setting a random controller for Transformer Block M is to ensure the overall feature output of real-time network state information. Furthermore, the reason for setting a random controller only for the outputs of Transformer Blocks M / 2 to M is that shallow features often have poor expressive power, and to avoid the feature forgetting problem inherent in linear structures, the output features of the tail Transformer Blocks are selectively acquired as part of the initial task features.

[0036] Step S2: By using the computing power scheduling network state awareness model, network state features are extracted from the above real-time network state information to generate network state features.

[0037] In practice, graph attention networks (GATs) can be used to create state-aware models for computing power scheduling networks. Specifically, because the network nodes in a computing power scheduling network have irregular connections, conventional neural networks based on Euclidean distance calculations are insufficient for effectively extracting the network state. Therefore, graph attention networks are employed to extract features from real-time network state information.

[0038] Step S3: Based on the above task characteristics and task type classifier, determine the task type corresponding to the above-mentioned computation task to be executed.

[0039] In practice, the task type classifier uses the FCN (Fully Convolutional Networks) model. The task type characterizes the type of computational task to be performed. Specifically, task types can include: load balancing task types, regular task types, and latency-priority task types.

[0040] Step S4: Match the computing power scheduling model corresponding to the above task types.

[0041] In practice, the computing power scheduling model can be an expert model. In particular, the computing power scheduling model can be a routing policy generator constructed using a hybrid expert model. Different task types have pre-set corresponding computing power scheduling models. Specifically, the computing power scheduling model adopts an encoder-decoder structure.

[0042] Step S5: Based on the above computing power scheduling model and the above task characteristics, generate the task execution description information included in the above task scheduling information.

[0043] In some embodiments, the aforementioned execution entity can input task characteristics into the computing power scheduling model to generate task execution description information.

[0044] Step S6: Determine the set of candidate routing paths.

[0045] The candidate routing path is a routing path that starts from the first network node and ends at the second network node. The first network node is a network node of type scheduling node that receives the aforementioned computational task to be executed. The second network node is a network node of type task execution node that corresponds to the aforementioned task execution description information.

[0046] In practice, when the starting point and the ending point are known, a graph traversal of the computing power scheduling network can be performed using a method such as depth-first traversal to obtain a set of candidate routing paths.

[0047] Step S7: Based on the above network state characteristics, determine the path score of each candidate route path in the above candidate route path set.

[0048] In practice, since network state features are network features specific to the entire computing power scheduling network, for each candidate routing path, firstly, network nodes other than those included in the candidate routing path are masked. Then, the features corresponding to these network nodes in the network state features are projected to zero, resulting in path state features for the candidate routing path. Next, the path state features are input into the FCN model (a model used to score candidate routing paths) to obtain the path score for each candidate routing path. By masking network nodes other than those included in the candidate routing path to locate the projection position, and further setting the features of the projection position to zero, the amount of subsequent feature processing is reduced.

[0049] Step S8: Generate the task route description information included in the above task scheduling information based on the target route path.

[0050] The target route path mentioned above refers to the candidate route path in the candidate route path set whose corresponding path score meets the path score filtering condition. The path score filtering condition is: the corresponding path score is the highest. The execution entity can use the target route path as task routing description information.

[0051] Step 103: Generate first heat dissipation control information based on task scheduling information, first node description information and pre-trained temperature prediction model.

[0052] In some embodiments, the aforementioned execution entity may generate first heat dissipation control information based on task scheduling information, first node description information, and a pre-trained temperature prediction model.

[0053] The first node description information represents the node temperature status of the network nodes involved in the task scheduling information before the computational task to be executed has started. Specifically, the first node description information includes the node temperature status of each network node involved in the task scheduling information. The node temperature status is stored in sequence. The temperature prediction model employs a recurrent neural network (RNN) model. The temperature prediction model takes the node temperature status of the network nodes included in the target routing path corresponding to the task routing description information as input and predicts the temperature status of the network nodes after a duration of k. Based on the predicted temperature status of the network nodes in the target routing path, a corresponding temperature control strategy is generated as the first heat dissipation control information.

[0054] In some optional implementations of certain embodiments, the execution entity generates first heat dissipation control information based on the task scheduling information, the first node description information, and the pre-trained temperature prediction model, including:

[0055] Step S1: Based on the task routing description information above, determine the predicted latency corresponding to each target network node in the target network node set.

[0056] The target network node set refers to the network nodes involved in the task scheduling information. Specifically, the target network node set can be network nodes on the target routing path.

[0057] In practice, given the target routing path corresponding to the known task routing description information, the arrival time of the computation task to be executed to each network node in the target network node set can be predicted. Specifically, using the time when the first target network node in the target network node set begins transmitting the computation task to be executed as the base time, the predicted delay for each network node to receive the computation task to be executed is determined by the time difference between the arrival time and the base time.

[0058] Step S2: For each target network node in the above target network node set, perform the following first heat dissipation control information generation step:

[0059] Step S21: Using the predicted delay corresponding to the target network node as the prediction increment, generate the predicted temperature corresponding to the target network node based on the first node description information and the temperature prediction model.

[0060] In practice, since the time it takes for the computation task to be executed to reach different network nodes (target network nodes) is different when it is transmitted along the target route path corresponding to the task route description information, there are different heat dissipation control prediction timings for the network nodes on the target route path, that is, different values ​​of k are corresponding to the network nodes on the target route path.

[0061] As an example, Figure 4 The diagram illustrates the target routing path, which includes n network nodes: A1, A2, ..., An-1, and An. The time it takes for network node A1 (a scheduling node) to receive the computation task to be executed is T; the time it takes for network node A2 to receive the task is T+Δk1; the time it takes for network node An-1 to receive the task is T+Δkn-1; and the time it takes for network node An to receive the task is T+Δkn. Therefore, the value of k for network node A2 is T+Δk1, the value of k for network node An-1 is T+Δkn-1, and the value of k for network node An is T+Δkn.

[0062] In practice, predicting the temperature after k steps using an RNN model combined with a temperature value sequence (the first node's description information) is a typical "rolling prediction problem." Because the prediction needs to be performed in a rolling manner based on the prediction results (predicted temperature), the accuracy may decrease due to the relatively long k steps. However, the temperature prediction involved in this disclosure mainly involves (Δk) as the k steps. Due to the efficiency of network transmission, the value of k is rarely very large, thus avoiding the aforementioned problem of reduced accuracy.

[0063] As another example, the temperature sequence for a network node represented by the first node description information could be [45, 47, 48, 50, 50, 49]. The k-step (Δk) value could be 10 seconds, with seconds as the time granularity. A temperature prediction model (RNN model) would then predict 10 values ​​in a rolling fashion, and the last predicted value would be used as the predicted temperature for the target network node. For example, first, [45, 47, 48, 50, 50, 49] would be input into the temperature prediction model to obtain the predicted temperature value A1. Then, [45, 47, 48, 50, 50, 49, predicted temperature value A1] would be input into the temperature prediction model to obtain the predicted temperature value A2, and so on, rolling the prediction for k steps to obtain the predicted temperature value Ak, which would then be used as the predicted temperature for the target network node.

[0064] Step S22: Determine the heat dissipation strategy information corresponding to the above target network node.

[0065] The heat dissipation strategy information represents the heat dissipation strategies of network nodes under different operating and environmental conditions. Specifically, the heat dissipation strategy information can include different heat dissipation rules. Different heat dissipation rules correspond to different heat dissipation modes. In practice, due to the different node types of network nodes, the corresponding heat dissipation strategies also differ. In practice, different heat dissipation strategy information can be pre-configured for different node types of network nodes. The aforementioned execution entity can determine the heat dissipation strategy information corresponding to the target network node through information matching.

[0066] Step S23: Based on the predicted temperature and heat dissipation strategy information corresponding to the target network node, determine the heat dissipation mode corresponding to the target network node.

[0067] In practice, since the heat dissipation strategy information includes different heat dissipation rules, the corresponding heat dissipation mode can be determined by predicting the temperature and matching the corresponding heat dissipation rules.

[0068] Step S24: Generate sub-heating control information based on the heat dissipation mode, predicted temperature and heat dissipation strategy information corresponding to the target network node.

[0069] In practice, the aforementioned implementing entity can determine the heat dissipation mode, predicted temperature, heat dissipation strategy information, and predicted latency corresponding to the target network node as sub-heat dissipation control information.

[0070] Step S3: Generate the first heat dissipation control information as described above based on the obtained set of sub-heat dissipation control information.

[0071] In practice, the aforementioned executing entity can obtain the first heat dissipation control information for the target routing path by using the set of sub-heat dissipation control information corresponding to the target network node set.

[0072] In some optional implementations of some embodiments, the above method further includes:

[0073] Step S4: In response to the completion of the above-mentioned computational task, adjust the target network nodes in the target network node set whose corresponding node type is scheduling node type or task execution node type to the first standard heat dissipation mode.

[0074] The first standard cooling mode is an active cooling mode, which controls the temperature through a constant-speed cooling method, such as a cooling fan. In practice, since target network nodes of the scheduling node type or task execution node type often correspond to a large number of computing task scheduling and execution, active cooling control is required to ensure that the network node can maintain a better temperature environment when no computing task scheduling or execution is being performed.

[0075] Step S5: In response to the completion of the above-mentioned computational task, adjust the target network node in the target network node set whose corresponding node type is routing node to the second standard heat dissipation mode.

[0076] The second standard cooling mode is a passive cooling mode, which uses a temperature threshold trigger to control temperature adjustment. In practice, since target network nodes of the routing node type are mainly used for routing and forwarding, and have relatively low data processing pressure, the temperature change of the node is small when it is not performing computational tasks for forwarding. Therefore, a passive cooling mode is used to control the ambient temperature when it is not performing computational tasks for forwarding.

[0077] Step 104: Based on the first heat dissipation control information, perform heat dissipation pre-control on the network nodes involved in the task scheduling information.

[0078] In some embodiments, the aforementioned execution entity may perform heat dissipation pre-control on the network nodes involved in the task scheduling information based on the first heat dissipation control information.

[0079] In practice, the aforementioned execution entity can control the target network nodes included in the target routing path to perform pre-heating based on the first heat dissipation control information, so as to ensure that the network nodes are at a suitable ambient temperature when receiving the computing task to be executed, so that the network nodes can process the computing task efficiently.

[0080] In some optional implementations of certain embodiments, pre-controlling the heat dissipation of the network nodes involved in the task scheduling information based on the first heat dissipation control information includes:

[0081] Step S1: For each target network node in the target network node set, perform the following heat dissipation pre-control steps:

[0082] Step S11: Generate the task listener corresponding to the target network node mentioned above.

[0083] Task listeners can be used to monitor the acceptance and forwarding status of pending computational tasks. In practice, process listeners can be created to serve as task listeners.

[0084] Step S12: Based on the task routing description information above, generate the heat dissipation initiation time for the target network node under the preset confidence level.

[0085] In practice, since the time (T+△k) at which each target routing node in the target network node set corresponding to the task routing description information receives the computation task to be executed is known, the heat dissipation initiation time under a preset confidence level can be calculated statistically. For example, the preset confidence level can be 90%.

[0086] Step S13: In response to the arrival of the heat dissipation initiation time, heat dissipation control is initiated for the target network node based on the sub-heat dissipation control information corresponding to the target network node.

[0087] In practice, the cooling devices corresponding to the target network node can be controlled to dissipate heat from the target network node using the sub-cooling control information. This ensures that the computational task to be executed is transmitted to the target network node in an optimal working environment.

[0088] In some optional implementations of some embodiments, the above method further includes:

[0089] Step S1: Collect the description information of the second node.

[0090] The second node description information described above represents the node temperature status of the network nodes involved in the task scheduling information when the computing task to be executed is being performed.

[0091] In practice, to ensure temperature control during the execution of computational tasks on the network nodes (target network nodes) involved in the aforementioned task scheduling information, it is possible to further collect second node description information in real time. Due to the timeliness requirement for information collection, the second node description information can be actively collected by the network node of the scheduling node type. Monitoring tools (e.g., Zabbix) can also be deployed in the computing power scheduling network, and a proxy method can be used to collect the node status information corresponding to the network nodes.

[0092] Step S2: Generate second heat dissipation control information based on the second node description information described above.

[0093] In practice, firstly, when the computational task to be executed is being performed, it mainly involves network nodes of the task execution node type (target network nodes). Therefore, the second heat dissipation control information is mainly for heat dissipation optimization of this type of network node. The generation method of the second heat dissipation control information is similar to that of the first heat dissipation control information.

[0094] In some optional implementations of certain embodiments, the execution entity generates second heat dissipation control information based on the second node description information, including:

[0095] Step S21: Select target network nodes that meet the filtering conditions from the above target network node set as network nodes to be optimized, and obtain the set of network nodes to be optimized.

[0096] The filtering condition is that the target network node's node type is the task execution node type.

[0097] Step S22: For each network node to be optimized in the above set of network nodes to be optimized, perform the following second heat dissipation control information generation step:

[0098] Step S221: Using a preset duration as the prediction increment, generate the updated predicted temperature corresponding to the above-mentioned network node to be optimized based on the description information of the second node and the temperature prediction model described above.

[0099] In practice, the preset duration corresponding to the target network node (the network node to be optimized) whose node type is task execution node is used as the prediction increment. Based on the second node description information and the temperature prediction model mentioned above, the predicted temperature corresponding to the target network node is generated. Specifically, the node temperature status of the target network node whose node type is task execution node, as described in the second node description information, is used as the model input to predict the predicted temperature after the preset duration. Since the target network node whose node type is task execution node has already started calculating the task execution at this time, the prediction latency is no longer used as the prediction increment, but the preset duration is used as the prediction increment. This ensures the matching degree of the prediction. Secondly, after obtaining the predicted temperature, the corresponding heat dissipation strategy information is further matched.

[0100] Step S222: In response to the temperature difference between the updated predicted temperature of the network node to be optimized and the predicted temperature of the network node to be optimized being greater than a preset temperature difference, determine the updated heat dissipation mode of the network node to be optimized based on the updated predicted temperature of the network node to be optimized and the heat dissipation strategy information.

[0101] In practice, when the temperature difference is greater than the preset temperature difference, it indicates a large temperature change. At this time, the corresponding heat dissipation strategy information changes, so it is necessary to update the corresponding heat dissipation mode as the updated heat dissipation mode.

[0102] Step S223: Generate updated sub-heating control information based on the updated heat dissipation mode, updated predicted temperature and heat dissipation strategy information corresponding to the target network node.

[0103] In practice, the updated heat dissipation mode, updated predicted temperature, heat dissipation strategy information, and preset duration corresponding to the target network node can be used as the updated sub-heat dissipation control information. Among them, the preset duration can be -1, indicating that heat dissipation is executed immediately.

[0104] Step S23: Generate the second heat dissipation control information as described above based on the obtained updated sub-heat dissipation control information set.

[0105] In practice, when distributed computing exists, multiple target network nodes of the task execution node type may be involved. Therefore, multiple updated sub-heating control information will be obtained. At this time, the set of updated sub-heating control information can be used as the second heat dissipation control information.

[0106] Step S3: Based on the second heat dissipation control information mentioned above, perform heat dissipation micro-control on the network nodes involved in the task scheduling information mentioned above.

[0107] In some embodiments, the aforementioned execution entity can perform heat dissipation micro-control on the network nodes involved in the aforementioned task scheduling information based on the aforementioned second heat dissipation control information. In practice, since the computing task to be executed has already started, there is no need to set the heat dissipation initiation time as in the heat dissipation pre-control stage. Instead, the second heat dissipation control information can be executed immediately to perform heat dissipation micro-control on the network nodes involved in the aforementioned task scheduling information, especially the target network nodes whose node type is the task execution node type.

[0108] The above embodiments of this disclosure have the following beneficial effects: Through the information processing and control method based on computing power scheduling in some embodiments of this disclosure, heat dissipation control under efficient computing power scheduling is achieved. Specifically, firstly, in response to the existence of a computing task to be executed, real-time network status information is collected. This real-time network status information represents the real-time network status of the computing power scheduling network. The node types corresponding to the network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type. This obtains the network communication status of the computing power scheduling network. Secondly, based on the task description information corresponding to the computing task to be executed and the real-time network status information, task scheduling information is generated. This task scheduling information includes: task routing description information and task execution description information. This generates task scheduling information matching the computing task to be executed from the perspective of task routing path and computing power allocation. Next, based on the task scheduling information, the first node description information, and a pre-trained temperature prediction model, first heat dissipation control information is generated. The first node description information represents the node temperature status of the network nodes involved in the task scheduling information when the computing task to be executed has not yet started execution. Finally, based on the aforementioned first heat dissipation control information, heat dissipation pre-control is performed on the network nodes involved in the aforementioned task scheduling information. In this way, heat dissipation pre-control of network nodes is achieved by combining task scheduling and node temperature status, thereby ensuring that network nodes execute computational tasks at optimal ambient temperatures, thus achieving the goal of heat dissipation control under efficient computing power scheduling.

[0109] Further reference Figure 5 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of an information processing and control device based on computing power scheduling. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this information processing and control device based on computing power scheduling can be specifically applied to various electronic devices.

[0110] like Figure 5As shown, some embodiments of the information processing and control device 500 based on computing power scheduling include: a data acquisition unit 501, a first generation unit 502, a second generation unit 503, and a heat dissipation pre-control unit 504. The acquisition unit 501 is configured to acquire real-time network status information in response to the existence of a computing task to be executed. The real-time network status information represents the real-time network status of the computing power scheduling network. The node types corresponding to the network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type. The first generation unit 502 is configured to generate task scheduling information based on the task description information corresponding to the computing task to be executed and the real-time network status information. The task scheduling information includes: task routing description information and task execution description information. The second generation unit 503 is configured to generate first heat dissipation control information based on the task scheduling information, the first node description information, and a pre-trained temperature prediction model. The first node description information represents the node temperature status of the network nodes involved in the task scheduling information when the computing task to be executed has not started execution. The heat dissipation pre-control unit 504 is configured to perform heat dissipation pre-control on the network nodes involved in the task scheduling information based on the first heat dissipation control information.

[0111] It is understandable that the units described in the computing power scheduling-based information processing and control device 500 are related to the reference... Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method also apply to the information processing and control device 500 based on computing power scheduling and the units contained therein, and will not be repeated here.

[0112] The following is for reference. Figure 6 It illustrates a schematic diagram of the structure of an electronic device (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of this disclosure. Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any of the methods described above. The processor provides computational and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the execution of the computer program in the non-volatile storage medium; when executed by the processor, the computer program causes the processor to perform any of the methods described above. The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 6The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0113] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0114] In one embodiment, the processor is configured to run a computer program stored in a memory to perform the following steps: In response to the existence of a computational task to be executed, real-time network status information is collected, wherein the real-time network status information characterizes the real-time network status of the computing power scheduling network, and the node types corresponding to network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type; task scheduling information is generated based on the task description information corresponding to the computational task to be executed and the real-time network status information, wherein the task scheduling information includes: task routing description information and task execution description information; first heat dissipation control information is generated based on the task scheduling information, first node description information, and a pre-trained temperature prediction model, wherein the first node description information characterizes the node temperature status of the network nodes involved in the task scheduling information when the computational task to be executed has not yet started execution; and heat dissipation pre-control is performed on the network nodes involved in the task scheduling information based on the first heat dissipation control information.

[0115] This disclosure also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the method implemented when the program instructions are executed can be referred to the various embodiments of the methods described above.

[0116] The aforementioned computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. Alternatively, the aforementioned computer-readable storage medium may be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0117] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0118] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. An information processing and control method based on computing power scheduling, characterized in that, include: In response to the existence of computing tasks to be executed, real-time network status information is collected, wherein the real-time network status information represents the real-time network status of the computing power scheduling network, and the node types corresponding to the network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type. Based on the task description information corresponding to the computation task to be executed and the real-time network status information, task scheduling information is generated, wherein the task scheduling information includes: task routing description information and task execution description information; Based on the task scheduling information, the first node description information, and the pre-trained temperature prediction model, first heat dissipation control information is generated, wherein the first node description information characterizes the node temperature state of the network node involved in the task scheduling information when the computing task to be executed has not started execution; Based on the first heat dissipation control information, heat dissipation pre-control is performed on the network nodes involved in the task scheduling information; The generation of task scheduling information includes: Task features are generated by extracting task features from the task description information using a task feature extraction model. By using a computing power scheduling network state awareness model, network state features are extracted from the real-time network state information to generate network state features. Based on the task characteristics and the task type classifier, the task type corresponding to the computation task to be executed is determined; Match the computing power scheduling model corresponding to the task type; Based on the computing power scheduling model and the task characteristics, the task scheduling information includes task execution description information; A set of candidate routing paths is determined, wherein the candidate routing path is a routing path with a first network node as the starting point and a second network node as the ending point. The first network node is a network node of type scheduling node that receives the computing task to be executed, and the second network node is a network node of type task execution node that corresponds to the task execution description information. Based on the network state characteristics, determine the path score for each candidate route path in the candidate route path set; Based on the target routing path, the task scheduling information includes task routing description information, wherein the candidate routing paths in the candidate routing path set are those whose corresponding path scores meet the path score filtering conditions.

2. The method of claim 1, wherein, The method further includes: Collect second node description information, wherein the second node description information represents the node temperature status of the network node involved in the task scheduling information when the computing task to be executed is executed; Based on the description information of the second node, generate the second heat dissipation control information; Based on the second heat dissipation control information, micro-control of heat dissipation is performed on the network nodes involved in the task scheduling information.

3. The method of claim 2, wherein, The step of generating first heat dissipation control information based on the task scheduling information, the first node description information, and the pre-trained temperature prediction model includes: Based on the task routing description information, the predicted latency corresponding to each target network node in the target network node set is determined, wherein the target network node set is the network nodes involved in the task scheduling information; For each target network node in the target network node set, the following first heat dissipation control information generation step is performed: Using the predicted delay corresponding to the target network node as the prediction increment, the predicted temperature corresponding to the target network node is generated based on the first node description information and the temperature prediction model. Determine the heat dissipation strategy information corresponding to the target network node; Based on the predicted temperature and heat dissipation strategy information corresponding to the target network node, the heat dissipation mode corresponding to the target network node is determined; Based on the heat dissipation mode, predicted temperature, and heat dissipation strategy information corresponding to the target network node, sub-heat dissipation control information is generated. The first heat dissipation control information is generated based on the obtained set of sub-heat dissipation control information.

4. The method of claim 3, wherein, The method further includes: In response to the completion of the computation task to be executed, the target network nodes in the target network node set whose corresponding node type is scheduling node type or task execution node type are adjusted to the first standard heat dissipation mode; In response to the completion of the computation task to be executed, the target network nodes in the target network node set whose corresponding node type is routing node are adjusted to the second standard heat dissipation mode.

5. The method of claim 4, wherein, The step of performing heat dissipation pre-control on the network nodes involved in the task scheduling information based on the first heat dissipation control information includes: For each target network node in the target network node set, perform the following heat dissipation pre-control steps: Generate a task listener corresponding to the target network node; Based on the task routing description information, a heat dissipation initiation time for the target network node is generated under a preset confidence level; In response to the arrival of the heat dissipation initiation time, heat dissipation control is initiated for the target network node based on the sub-heat dissipation control information corresponding to the target network node; In response to the task listener detecting that the target network node has completed the task execution for the computing task to be executed, the heat dissipation control of the target network node based on the sub-heat dissipation control information corresponding to the target network node is stopped.

6. The method of claim 5, wherein, The step of generating second heat dissipation control information based on the second node description information includes: From the target network node set, target network nodes that meet the filtering conditions are selected as network nodes to be optimized, thus obtaining the set of network nodes to be optimized; For each network node to be optimized in the set of network nodes to be optimized, the following second heat dissipation control information generation step is performed: Using a preset duration as the prediction increment, the updated predicted temperature corresponding to the network node to be optimized is generated based on the description information of the second node and the temperature prediction model. In response to the temperature difference between the updated predicted temperature of the network node to be optimized and the predicted temperature of the network node to be optimized being greater than a preset temperature difference, the updated heat dissipation mode of the network node to be optimized is determined based on the updated predicted temperature of the network node to be optimized and the heat dissipation strategy information. Based on the updated heat dissipation mode, updated predicted temperature and heat dissipation strategy information corresponding to the target network node, updated sub-heat dissipation control information is generated. The second heat dissipation control information is generated based on the updated set of sub-heat dissipation control information.

7. A computing power scheduling-based information processing and control device, characterized by comprising: include: The acquisition unit is configured to acquire real-time network status information in response to the existence of a computing task to be executed. The real-time network status information represents the real-time network status of the computing power scheduling network. The node types corresponding to the network nodes in the computing power scheduling network include: scheduling node type, routing node type, and task execution node type. The first generation unit is configured to generate task scheduling information based on the task description information corresponding to the computation task to be executed and the real-time network status information, wherein the task scheduling information includes: task routing description information and task execution description information; wherein generating the task scheduling information includes: Task features are generated by extracting task features from the task description information using a task feature extraction model. By using a computing power scheduling network state awareness model, network state features are extracted from the real-time network state information to generate network state features. Based on the task characteristics and the task type classifier, the task type corresponding to the computation task to be executed is determined; Match the computing power scheduling model corresponding to the task type; Based on the computing power scheduling model and the task characteristics, the task scheduling information includes task execution description information; A set of candidate routing paths is determined, wherein the candidate routing path is a routing path with a first network node as the starting point and a second network node as the ending point. The first network node is a network node of type scheduling node that receives the computing task to be executed, and the second network node is a network node of type task execution node that corresponds to the task execution description information. Based on the network state characteristics, determine the path score for each candidate route path in the candidate route path set; Based on the target routing path, the task scheduling information includes task routing description information, wherein the target routing path is a candidate routing path in the candidate routing path set whose corresponding path score meets the path score filtering condition; The second generation unit is configured to generate first heat dissipation control information based on the task scheduling information, the first node description information and the pre-trained temperature prediction model, wherein the first node description information characterizes the node temperature state of the network node involved in the task scheduling information when the computing task to be executed has not started execution. The heat dissipation pre-control unit is configured to perform heat dissipation pre-control on the network nodes involved in the task scheduling information based on the first heat dissipation control information.

8. An electronic device, comprising: include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

9. A computer readable medium characterized by It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images