CN110730138A - Dynamic resource allocation method, system and storage medium for space-based cloud computing architecture - Google Patents

Abstract

The present invention provides a dynamic resource allocation method, system and storage medium under a space-based cloud-fog computing architecture. The method includes: sensing the state parameters of the current network, and generating a network state sensing result; and detecting, according to the network state sensing result, the traffic situation of each node and link, the amount of user visits, and each application under the current network topology structure. Calculate the consumption of traffic, and classify and identify the above traffic to generate traffic classification and identification results; according to the traffic classification and identification results, predict the traffic situation based on geographic location, time distribution, and access density, and generate traffic prediction results; According to the traffic forecast results, the resource allocation plan is generated visually through the big data analysis platform. The invention can efficiently utilize heterogeneous resources through dynamic resource allocation, meet the requirements of space-based delay sensitivity and big data applications, adapt to dynamic network connections, ensure the reliability of service flows, and achieve system load balance.

Description

Dynamic resource allocation method, system and storage medium for space-based cloud and fog computing architecture

technical field

The invention relates to the field of computer technology, and in particular, to a method, system and storage medium for dynamic resource allocation of a space-based cloud-fog computing architecture.

Background technique

With the development of technology and the diversification of application requirements, the traditional "chimney-type" satellite technology has a single function, is isolated from each other, and its operation is heavily dependent on the ground. Problems such as these are prominent, which cannot meet the development of my country's military, economy and people's living standards. urgent needs. The integrated information network of space and earth comprehensively utilizes new information network technologies and gives full play to the advantages of space-based and ground-based networks, which can support major national strategic actions and promote the transmission and sharing of diverse information. Satellite network is an important part of the integrated information network of space and earth, which has the advantages of high, long distance and wide coverage.

The major project of my country's Innovation 2030 "Integrated Information Network of Heaven and Earth" aims to build an information network system with global coverage and on-demand services. In the major project of the integrated information network of the sky and the ground, the "sky-net-ground network" architecture is adopted. Mobile communication network interconnection. The space-based backbone network is mainly composed of space-based backbone nodes deployed in the GEO orbit through high-speed interconnection of inter-satellite links, with global coverage capabilities; the space-based access network is mainly composed of constellations deployed in the LEO orbit, with global seamless It is also known as the space-based access network low-orbit constellation. The satellite optical network based on space laser communication has the advantages of high speed and large information capacity, and is an important technical means for satellite relay and satellite networking in the future.

With the continuous improvement of the concept of the integrated network between the sky and the ground, combined with the development of terrestrial networking technologies, especially the emergence of technologies such as Software Defined Network (SDN)/Network Functions Virtualization (NFV), EC, etc. The rapid development of artificial intelligence (AI) technology in recent years has made the satellite network gradually develop towards intelligence, coordination and identification, and the transition from "host-centric" to "information-centric". trend.

The research on space-based network using emerging technologies of ground-based network mainly includes: research on integrated space-ground network based on SDN/NFV/MEC, research on integration of satellite network and terrestrial 5G network, architecture design of satellite network resource management based on AI, integrated space-ground network Protocol research, as well as research on satellite network links. SDN/NFV/MEC are the three key technologies of 5G mobile communication, and AI technology plays an important role in the application of network intelligence. It can be seen that the research on emerging network technologies is not only conducive to promoting the development of terrestrial Internet and mobile communication networks It also plays a pivotal role in the construction of the integrated information network of heaven and earth.

At the same time of theoretical research and simulation realization, the research and development of experimental prototype are also in progress. On November 20, 2018, my country's first experimental satellite "Tianzhi-1", which was developed by the Institute of Software of the Chinese Academy of Sciences, was successfully launched at the Jiuquan Satellite Launch Center. The successful launch of Tianzhi satellite provides an open test platform for the development of space-based intelligence, provides assistance for commercial aerospace development and aerospace ecosystem construction, and promotes the evolution of traditional satellites in the direction of platformization, softwareization, intelligence, and virtualization. .

In the Cloud Computing (CC) mode, the User Equipment (UE) can access and utilize the computing and storage resources of the highly powerful remote centralized CC center through the operator and the Core Network (CN). As shown in Figure 1. In recent years, the number of UEs accessing the Internet has exploded, and in the new Internet of Things (IoT) model, countless heterogeneous devices with extensive computing capabilities will be interconnected. This also means that the network load will show an increasing trend, and the linear growth of centralized CC capabilities cannot match the exponential growth of massive edge data. The massive amount of data transmitted from edge UEs to the CC center increases the load of network transmission bandwidth and introduces high latency. Moreover, the data transmission causes a large power consumption of the UE with limited energy.

In order to solve the problem of high network load and meet users' requirements for ultra-low latency and ultra-high bandwidth, CC services should be moved to the vicinity of the UE. In terms of network topology, that is, the edge of the network, just like the emerging edge computing ( EdgeComputing, EC) paradigm considered. These EC paradigms place computing/storage capabilities at the network edge, such as Cloudlet, mobile ad-hoc cloud, Fog Computing (FC), Mobile Edge Computing/Multi-access Edge Computing , MEC), has the advantages of achieving lower latency, saving UE energy consumption, supporting real-time information on radio networks and location-aware computing, alleviating network congestion, and enhancing the privacy and security of mobile applications.

With the continuous development of the aviation industry, the number of satellites in orbit with various functions continues to increase, generating massive amounts of real-time data. Especially in remote sensing data analysis, GEO satellites and medium and low orbit satellites each collect remote sensing data information, which produces a large amount of real-time data. Under normal circumstances, these data need to be downloaded to the ground station for processing, that is, the CC mode is adopted. This resource-intensive big data processing method can provide fast data processing capabilities and create economic benefits of effective scale. However, similar to the terrestrial network, this approach exposes the same problems with the growing number of devices in orbit and the growing network traffic.

This requires us to add computing power to the on-orbit device side, so that its data can be processed in real time, so as to avoid affecting its function due to the limitation of network bandwidth. The application of EC technology is on the edge side of the network, an open platform that integrates network, computing, storage, and application core capabilities, and provides edge intelligent processing services nearby to meet the needs of satellite equipment in business implementation, data optimization, security and privacy protection. key needs. In view of the huge amount of data processed on-orbit, the combination of space-based edge computing technology is conducive to improving the ability of remote sensing data on-orbit processing, realizing on-orbit real-time processing, rapid distribution on demand, improving real-time decision-making and reducing processing time.

In the research of space-based computing architecture, in order to solve the contradiction between time-sensitive and big-data space applications and limited satellite-ground bandwidth, with the support of technologies such as software-defined satellites, virtualization, and space networks, there has been a A cloud-based fog computing system is proposed. The system mainly includes: space edge node, space edge cloud and ground remote cloud.

Space edge nodes can be divided into user nodes (satellites or aircraft) and fog satellite nodes. The user node does not undertake or only undertakes a relatively small amount of computing functions, and mainly offloads services to the edge cloud or remote cloud to complete. The fog satellite node has certain computing and storage capabilities. It can be used as a user to offload its computing tasks to the edge cloud or remote cloud to request assistance for completion, and it can also be used as a computing resource to accept tasks from the cloud and independently or Perform tasks by forming a fast service cluster with other fog satellite nodes. Fog satellite nodes use a general virtualization platform and can carry different types of services according to requirements.

The space edge cloud is the infrastructure of the space-based information port, which assumes the function of the space-based edge data center. Compared with fog satellites, space-based edge clouds integrate CPU, GPU, FPGA and other heterogeneous resources, and have more powerful computing and storage capabilities. With the support of virtualization technology, it can not only complete various applications offloaded from user nodes, but also complete functions such as data fusion, task analysis, intelligent distribution, and building a fast service cluster of fog satellite nodes.

The remote cloud on the ground is the infrastructure of the ground-based information port, which undertakes the function of a large-scale cloud computing center. Compared with the space-based edge cloud, the terrestrial cloud has more computing storage resources and stronger computing power.

Fog satellite nodes, spatial edge clouds, and ground remote clouds constitute the three-layer computing model of the spatial information fog network. From the edge to the cloud, the computing power of each layer is gradually enhanced. From the point of view of computing allocation, computing with low complexity can be completed on fog satellite nodes; computing with higher complexity and higher real-time processing requirements is suitable for completion on the space-based edge cloud; Calculations with low requirements, large amount of calculation and high computational complexity should be completed on the ground remote cloud.

The particularity of space-based computing resources is as follows: (1) Heterogeneity: computing resources on satellites include CPU, FPGA, GPU, memory, etc.; (2) Dispersion: satellite computing resources are scattered in various locations in space; (3) ) Dynamics: the satellite is in motion, and the topology of the spatial information network is time-varying.

Therefore, designing a dynamic resource allocation method process adapted to the space-based cloud-fog computing architecture, and how to configure the computing/storage resources on the satellite to achieve the corresponding target requirements, are the key and difficult points worth studying.

SUMMARY OF THE INVENTION

In order to solve at least one of the above technical problems, the present invention provides a dynamic resource allocation method, system and storage medium of a space-based cloud-fog computing architecture.

In order to achieve the above object, the first aspect of the present invention proposes a dynamic resource allocation method for a space-based cloud-fog computing architecture. The dynamic resource allocation method includes:

Use existing network resources and network historical data to perceive the current network state parameters, and generate network state perception results;

According to the network status perception result, detect the traffic situation of each node and link, the amount of user access, and the traffic consumption of each application under the current network topology structure, and classify and identify the above traffic to generate traffic classification and identification results. ;

According to the traffic classification and identification results, summarize and summarize the traffic characteristics of each node and link under the current network topology, predict the traffic situation based on geographic location, time distribution, and access density, and generate traffic prediction results;

According to the traffic forecast results, the resource allocation plan is generated visually through the big data analysis platform.

In this solution, the state parameters of the current network are sensed by using existing network resources and network historical data, and also includes:

The state parameters of the current network are sensed through in-band network telemetry, where the state parameters are one or more of network physical topology, queue capacity, and single-hop delay.

In this solution, under the current network topology, the traffic conditions of each node and link, the amount of user access, and the consumption of traffic by each application are detected, and the above traffic is classified and identified, including:

The method of indexing the data stream by the hash function is used to detect the passing elephant stream online at high speed. When the data stream arrives, the scale of the data stream is quickly counted by the hash index, and the data stream whose size exceeds a certain threshold is determined as Elephant Stream; and/or

Based on traffic characteristics, the traffic is classified and identified by machine learning methods.

Further, based on the traffic characteristics, the traffic is classified and identified by the machine learning method, which also includes:

Define the characteristics by which traffic can be identified and differentiated;

An ML classifier is trained that associates the traffic's feature set with known classes, and the ML algorithm is applied to classify unknown traffic according to previously learned and trained rule models.

In this solution, according to the results of traffic classification and identification, the traffic characteristics of each node and link under the current network topology are summarized, and the traffic situation based on geographical location, time distribution and access density is predicted, including:

Through regression learning method and/or reinforcement learning method, the traffic situation based on geographic location, time distribution, and access density is predicted.

Preferably, the resources include computing resources and storage resources, the computing resources are one or more of CPU, GPU, and FPGA, and the storage resources are one or more of a mechanical hard disk, a solid-state hard disk, a liquid hard disk, and an optical hard disk. several.

The second aspect of the present invention also provides a dynamic resource allocation system of a space-based cloud and fog computing architecture. The dynamic resource allocation system of the space-based cloud and fog computing architecture includes: a memory and a processor, and the memory includes a space-based cloud and fog computing system. The dynamic resource allocation method program of the architecture, the dynamic resource allocation method program of the space-based cloud-fog computing architecture is executed by the processor to implement the following steps:

Use existing network resources and network historical data to perceive the current network state parameters, and generate network state perception results;

According to the network status perception result, detect the traffic situation of each node and link, the amount of user access, and the traffic consumption of each application under the current network topology structure, and classify and identify the above traffic to generate traffic classification and identification results. ;

According to the traffic classification and identification results, summarize and summarize the traffic characteristics of each node and link under the current network topology, predict the traffic situation based on geographic location, time distribution, and access density, and generate traffic prediction results;

According to the traffic forecast results, the resource allocation plan is generated visually through the big data analysis platform.

In this solution, the state parameters of the current network are sensed by using existing network resources and network historical data, and also includes:

The state parameters of the current network are sensed through in-band network telemetry, where the state parameters are one or more of network physical topology, queue capacity, and single-hop delay.

In this solution, under the current network topology, the traffic conditions of each node and link, the amount of user access, and the consumption of traffic by each application are detected, and the above traffic is classified and identified, including:

The method of indexing the data stream by the hash function is used to detect the passing elephant stream online at high speed. When the data stream arrives, the scale of the data stream is quickly counted by the hash index, and the data stream whose size exceeds a certain threshold is determined as Elephant Stream; and/or

Based on traffic characteristics, the traffic is classified and identified by machine learning methods.

A third aspect of the present invention further provides a computer-readable storage medium, the computer-readable storage medium includes a dynamic resource allocation method program of a space-based cloud and fog computing architecture, and the dynamic resource allocation method of the space-based cloud and fog computing architecture When the program is executed by the processor, the steps of the dynamic resource allocation method of the space-based cloud-fog computing architecture as described above are implemented.

By dynamically configuring computing resources and storage resources, the present invention can meet the requirements of 5G networks, achieve seamless global coverage, and can support an increase in the number of user connections to 1 million users per square kilometer. While achieving seamless global coverage, the invention further optimizes resource allocation, so that each resource node does not interfere with each other, reduces resource allocation cost, and maximizes resource utilization under the premise of limited resource allocation. By introducing edge computing, the invention enables the satellite to have on-orbit computing capability, shortens the propagation delay, and increases the propagation delay to the millisecond level, thereby ensuring that the data can be calculated and processed in real time.

The invention also uses high-throughput communication technologies such as laser communication to increase the amount of data transmitted, and at the same time, the satellite on-orbit calculation can occupy less bandwidth for transmission, and enable limited bandwidth to transmit more effective data. In addition, after the resources are initially allocated, according to the user service quality evaluation, the present invention can adjust the resources again, and continuously optimize the resource allocation.

Additional aspects and advantages of the present invention will be presented in the description which follows, in part, which will become apparent from the following description, or may be learned by practice of the invention.

Description of drawings

1 shows a network architecture diagram of a cloud computing mode in the prior art;

FIG. 2 shows the workflow of the resource allocation system under a space-based cloud-fog computing architecture of the present invention;

Fig. 3 shows the flow chart of the dynamic resource allocation method of a space-based cloud-fog computing architecture of the present invention;

4 shows a flowchart of a dynamic resource configuration method according to an embodiment of the present invention;

Fig. 5 shows a kind of in-band network monitoring architecture diagram of the present invention;

Fig. 6 shows the flow chart of a kind of elephant flow detection and prediction method of the present invention;

FIG. 7 shows a block diagram of a dynamic resource allocation system under a space-based cloud-fog computing architecture of the present invention.

Detailed ways

In order to understand the above objects, features and advantages of the present invention more clearly, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments may be combined with each other in the case of no conflict.

Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific details disclosed below. Example limitations.

As shown in Figure 2, the workflow of the resource allocation system under the space-based cloud-fog computing architecture includes: basic resource allocation, dynamic resource allocation, and satellite constellation design. The basic resource configuration includes the computing/storage resources that all nodes in the space-based cloud-fog computing architecture will configure in the underlying hardware, which can meet the most basic tasks. In general, the amount of resources required for this task will not be particularly large. Dynamic resource configuration includes some special computing/storage resources, computing resources can be heterogeneous computing resources (such as: GPU, CPU, FPGA), storage resources can be different storage resources (such as: mechanical hard disk, solid state hard disk, liquid hard disk, optical hard disk). Due to the high dynamic characteristics of the space-based network, it is necessary to design the corresponding satellite constellation orbits to complete the resource allocation according to the configuration requirements of the resource allocation system under the space-based cloud and fog computing architecture.

FIG. 3 shows a flow chart of a dynamic resource allocation method of a space-based cloud-fog computing architecture according to the present invention.

As shown in FIG. 3 , a first aspect of the present invention proposes a dynamic resource allocation method for a space-based cloud-fog computing architecture, and the dynamic resource allocation method includes:

S302, using existing network resources and network historical data to perceive the state parameters of the current network, and generate a network state perception result;

S304, according to the network state perception result, detect the traffic situation of each node and link, the user access amount, and the traffic consumption of each application under the current network topology structure, and classify and identify the above traffic to generate a traffic classification identification results;

S306, according to the traffic classification and identification results, summarize and summarize the traffic characteristics of each node and link under the current network topology structure, predict the traffic situation based on geographic location, time distribution, and access density, and generate a traffic prediction result;

S308 , according to the traffic prediction result, a resource allocation scheme is visually generated through the big data analysis platform.

It should be noted that the resources include computing resources and storage resources. The computing resources can be one or more of CPU, GPU, and FPGA, and the storage resources can be mechanical hard disks, solid-state hard disks, liquid hard disks, and optical hard disks. one or more of them.

It should be noted that the above dynamic resource allocation method adopts the logic of "network state perception + traffic identification + traffic prediction + generation of resource allocation scheme". Network status awareness mainly uses existing network resources and network historical data to perceive the current network status parameters. The main parameters include network topology, link traffic, link delay, node user traffic, and access network side traffic. Preferably, the network state awareness can be used to sense the entire network state through in-band network telemetry.

Traffic identification is mainly based on the results of network status perception, and detects the traffic conditions of each node and link (such as elephant flow), user access, and traffic consumption of each application under the current network topology. classification identification. Preferably, traffic identification can be performed through data stream indexing/machine learning technology.

Traffic prediction is mainly based on the results of traffic classification and identification, summarizes the traffic characteristics of each node and link under the current network topology, and predicts the traffic situation based on geographic location, time distribution, and access density. Preferably, traffic prediction can be performed by regression learning/reinforcement learning.

The generation of resource allocation plans is mainly based on the results of traffic forecasting, through the big data analysis platform to visually generate resource allocation plans, including resource allocation in geographic or spatial locations, based on services in different time periods (such as day and night or throughout the year) Periodic resource configuration for migration. Preferably, the generation of the resource allocation scheme may be based on a big data analysis platform that jointly generates the results of network status perception, traffic identification and traffic prediction.

As shown in Figure 4, through network state perception, traffic identification, and traffic prediction, a resource allocation plan is finally generated. If the resource allocation scheme can meet the target requirements, the satellite constellation can be designed according to the resource allocation scheme. If the target requirements cannot be met, the network can be re-perceived, traffic identified, and traffic predicted, and then a resource allocation plan can be generated and evaluated to form a closed-loop system, and iteratively update the resource allocation plan based on historical network data to continuously optimize the resource allocation plan.

According to the embodiment of the present invention, the state parameters of the current network are sensed by using existing network resources and network historical data, further comprising:

The state parameters of the current network are sensed through in-band network telemetry, where the state parameters are one or more of network physical topology, queue capacity, and single-hop delay.

It should be noted that the present invention uses the SDN-based in-band network telemetry technology to perceive the performance of nodes and links in the entire network, and specifically to perform detailed information on the status information of network devices (including network physical topology, queue capacity, single-hop delay, etc.). Granular and efficient monitoring. The present invention is based on the existing programmable network hardware plane, and intends to use a set of new network hardware monitoring system—in-band network perception technology to realize network real-time perception without hardware dependence, no extra traffic, packet level, and millisecond level perception. surveillance technology. The in-band network awareness technology is the network awareness technology of the data plane. The core idea is to directly write the network status data into the header of the data packet through programmable hardware.

As shown in Figure 5, the SDN controller only needs to issue monitoring instructions to the network equipment. When the network traffic passes through the node, the monitoring instruction is directly written in the data packet header through the programmable network device. When forwarding along the forwarding path, monitoring data will be continuously added to the packet header. Finally, at the end node, the telemetry data will be complexly separated from the network, and the telemetry data will be directly uploaded to the data analysis platform.

In-band network monitoring directly collects data from the data plane without the participation of the controller. The network status is directly written in the data packet header, and the monitoring data is directly uploaded to the big data analysis platform through the standard message queue. In this way, fine-grained network monitoring at the packet level is realized, while avoiding massive probe communication and the computational pressure of the controller. The above monitoring technology has the following advantages: (1) No additional hardware dependencies: INT does not need to rely on specific network hardware devices. For devices from different manufacturers, there is no hardware compatibility problem, which is conducive to promotion. (2) No additional traffic: INT writes the sensed network status data to the packet header of the data packet, instead of generating additional network status data packets. The amount of data is very small, and basically no additional traffic is generated in the network, which is conducive to maintaining the network. link stability. (3) Packet-level awareness: INT can perform network awareness at the packet level, and can obtain fine-grained status information such as delay and congestion of each packet. (4) Millisecond-level perception: Different from traditional network monitoring technologies based on sampling or statistical methods, INT can actively transmit network status through each data packet, and transmit data efficiently for applications such as network analysis in real time, which is suitable for delay-sensitive applications. Sexual business perception.

According to the embodiment of the present invention, under the current network topology, the traffic conditions of each node and link, the amount of user access, and the traffic consumption of each application are detected, and the above traffic is classified and identified, and further includes:

The method of indexing the data stream by the hash function is used to detect the passing elephant stream online at high speed. When the data stream arrives, the scale of the data stream is quickly counted by the hash index, and the data stream whose size exceeds a certain threshold is determined as Elephant Stream; and/or

Based on traffic characteristics, the traffic is classified and identified by machine learning methods.

Further, based on the traffic characteristics, the traffic is classified and identified by the machine learning method, which also includes:

Define the characteristics by which traffic can be identified and differentiated;

An ML classifier is trained that associates the traffic's feature set with known classes, and the ML algorithm is applied to classify unknown traffic according to previously learned and trained rule models.

It is understandable that big data has brought huge challenges to communication, and its characteristics of large scale, fast speed and variety are also fully reflected in communication. For example, in the mobile backbone network, the traffic bandwidth is 40G to 100Gbps; in a large data center, the traffic bandwidth can reach a scale of 1Tbps. In a communication network, a large-scale data flow is called an elephant flow, and a small-scale data flow is called a mouse flow. Although the proportion of elephant flow is not high, it will occupy network bandwidth and cause network congestion. Therefore, timely detection and prediction of elephant flows and taking necessary measures are important issues in network management.

It should be noted that the elephant flow detection means that when the data flow continuously passes through, the size of the data flow is counted to determine whether it is an elephant flow (that is, there are a large number of data packets). The invention adopts the method of indexing the data stream by the hash function to detect the passing elephant stream online at high speed. Use the hash function to index the data stream. When the data stream arrives, use the hash index to quickly count the size of the data stream. The data stream whose size exceeds a certain threshold is an elephant stream; in this process, the small-scale data The stream is dropped to prevent data overflow.

Further, the present invention also provides a traffic feature-based classification method, which performs traffic classification by identifying statistical patterns in externally observable attributes of traffic, without the need to deeply inspect the content of data packets to collect information and infer semantics. The method can ultimately cluster flows in the network into groups with similar traffic patterns, or one or more related application categories. Traffic characteristics include: packet inter-arrival time (average, variance, etc.), packet size (maximum, minimum, average), total bytes of the flow, duration of the flow, etc. In order to better integrate these traffic characteristics, a machine learning (ML) algorithm is preferably used for traffic classification.

The present invention applies machine learning technology to traffic classification based on traffic characteristics, which involves the following steps: First, it is necessary to define the characteristics that can be identified and differentiated by traffic, and these characteristics are traffic attributes calculated from multiple data packets (such as maximum or minimum packet length in each direction, traffic duration and packet inter-arrival time, etc.); then, an ML classifier is trained that can associate the traffic's feature set with known classes (created according to business needs) , and apply ML algorithms to classify unknown traffic using previously learned and trained rule models.

To be clear, each ML algorithm has different ways of sorting and optimizing feature sets, so different ML algorithms will have different dynamic behaviors during training and classification. The criteria for measuring traffic classification are: False Negatives (FN), False Positives (FP), True Positives (TP) and True Negatives (TN), recall, precision, traffic precision and byte precision, etc. When using ML for traffic classification, supervised learning and unsupervised learning are usually used, preferably using supervised learning algorithms such as Nearest Neighbor (NN), Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Naive Bayes, Genetic Algorithm, etc. And unsupervised learning algorithms such as Expectation Maximization (EM), AutoClass, K-Means, etc. for traffic classification. It is understandable that different ML algorithms have differences in classification accuracy, modeling time, classification speed, etc.

It should be noted that when applying machine learning to traffic classification, when classifying traffic in combination with specific services, factors such as packet loss, delay jitter, packet fragmentation, unknown flow direction, and CPU and memory resource consumption need to be considered. For ML algorithm classification performance impact.

Further, according to the traffic classification and identification results, summarize and summarize the traffic characteristics of each node and link under the current network topology structure, and predict the traffic situation based on geographic location, time distribution, and access density, including:

Through regression learning method and/or reinforcement learning method, the traffic situation based on geographic location, time distribution, and access density is predicted.

It should be noted that the elephant flow prediction means that when the data flow first arrives, the system can judge whether it is an elephant flow by only looking at its packet header information. Further, elephant flow prediction can be defined as an online regression learning problem, where a data flow is a sample consisting of a feature vector and the scale of the flow, as shown in Figure 6. The feature vector represents the address, time and other information of the data stream. The online learning system continuously learns the regression model from the data, and then predicts its size based on the features for a newly given sample. Preferably, the above model belongs to Gaussian Process Regression, the model is learned online from streaming data, the learned function follows a Gaussian process, and the samples to be predicted are given by the joint Gaussian distribution of the training samples and the samples to be predicted. The expectation and variance of , i.e. make predictions.

It can be understood that in artificial intelligence traffic prediction, network status awareness provides a wealth of fine-grained network status and traffic characteristics, and realizes global monitoring and perception of the entire network. However, with the expansion of the network scale and the increase in the number of services, the link will face a massive network state space, requiring agents to predict the state of traffic from the massive state space, which brings great difficulties to traditional machine learning. . The deep reinforcement learning algorithm of the present invention combines deep learning on the basis of reinforcement learning to compress the state space of reinforcement learning. Deep neural network has strong function fitting ability. Deep learning can not only bring the convenience of end-to-end optimization to reinforcement learning, but also make reinforcement learning no longer limited to low-dimensional space, which greatly expands the scope of reinforcement learning. Reinforcement learning defines the goal of optimization, and deep learning gives the operating mechanism - the way to represent the problem and the way to solve it. Therefore, the combination of reinforcement learning and deep learning enables accurate prediction of flow states from massive state spaces.

It can be understood that the real-time information of the network is transmitted to the AI plane through the controller, and is used as the input state (State) of the AI plane. Since the controller gets a global view of the network, the state space of the network input states is large. The agent through deep reinforcement learning takes the state as input and compresses the network state space. Deep reinforcement learning obtains the agent's policy by training the observed network data as state input. Deep reinforcement learning uses a policy (Policy) π=(a|s) to realize the mapping between the state space and the action space, and select a near-optimal decision action (Action) according to the input state. The AI plane feeds back the obtained decision-making actions to the controller and sends it to the underlying network for configuration and deployment through the controller. Network policy learning through deep reinforcement learning has the following three advantages:

First, the black-box characteristics of deep reinforcement learning algorithms enable different network decision-making tasks and optimization goals to only need to design action spaces and rewards without redesigning mathematical models;

Second, the powerful fitting ability of the deep neural network can process complex environments and find the optimal network strategy from the network state of the massive state space;

Third, once the agent of the deep reinforcement learning algorithm is trained, it can give an approximate optimal solution in one step, which has great advantages for highly dynamic networks compared to the multi-step convergence of heuristic algorithms.

FIG. 7 shows a block diagram of a dynamic resource allocation system under a space-based cloud-fog computing architecture of the present invention.

As shown in FIG. 7 , the second aspect of the present invention further proposes a dynamic resource allocation system 7 of a space-based cloud and fog computing architecture. The dynamic resource allocation system 7 of the space-based cloud and fog computing architecture includes: a memory 71 and a processor 72, so The memory 71 includes a dynamic resource allocation method program of a space-based cloud and fog computing architecture, and the dynamic resource allocation method program of the space-based cloud and fog computing architecture is executed by the processor 72. The following steps are implemented:

Use existing network resources and network historical data to perceive the current network state parameters, and generate network state perception results;

According to the network status perception result, detect the traffic situation of each node and link, the amount of user access, and the traffic consumption of each application under the current network topology structure, and classify and identify the above traffic to generate traffic classification and identification results. ;

According to the traffic classification and identification results, summarize and summarize the traffic characteristics of each node and link under the current network topology, predict the traffic situation based on geographic location, time distribution, and access density, and generate traffic prediction results;

According to the traffic forecast results, the resource allocation plan is generated visually through the big data analysis platform.

According to the embodiment of the present invention, the state parameters of the current network are sensed by using existing network resources and network historical data, further comprising:

The state parameters of the current network are sensed through in-band network telemetry, where the state parameters are one or more of network physical topology, queue capacity, and single-hop delay.

According to the embodiment of the present invention, under the current network topology, the traffic conditions of each node and link, the amount of user access, and the traffic consumption of each application are detected, and the above traffic is classified and identified, and further includes:

The method of indexing the data stream by the hash function is used to detect the passing elephant stream online at high speed. When the data stream arrives, the scale of the data stream is quickly counted by the hash index, and the data stream whose size exceeds a certain threshold is determined as Elephant Stream; and/or

Based on traffic characteristics, the traffic is classified and identified by machine learning methods.

A third aspect of the present invention further provides a computer-readable storage medium, the computer-readable storage medium includes a dynamic resource allocation method program of a space-based cloud and fog computing architecture, and the dynamic resource allocation method of the space-based cloud and fog computing architecture When the program is executed by the processor, the steps of the dynamic resource allocation method of the space-based cloud-fog computing architecture as described above are implemented.

By dynamically configuring computing resources and storage resources, the present invention can meet the requirements of 5G networks, achieve seamless global coverage, and can support an increase in the number of user connections to 1 million users per square kilometer. While achieving seamless global coverage, the invention further optimizes resource allocation, so that each resource node does not interfere with each other, reduces resource allocation cost, and maximizes resource utilization under the premise of limited resource allocation. By introducing edge computing, the invention enables the satellite to have on-orbit computing capability, shortens the propagation delay, and increases the propagation delay to the millisecond level, thereby ensuring that the data can be calculated and processed in real time.

The invention also uses high-throughput communication technologies such as laser communication to increase the amount of data transmitted, and at the same time, the satellite on-orbit calculation can occupy less bandwidth for transmission, and enable limited bandwidth to transmit more effective data. In addition, after the resources are initially allocated, according to the user service quality evaluation, the present invention can adjust the resources again, and continuously optimize the resource allocation.

The invention can efficiently utilize heterogeneous resources, meet the requirements of space-based time delay sensitivity and big data applications, adapt to dynamic network connections, ensure the reliability of service flows, and realize system load balance.

In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other manners. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined, or Can be integrated into another system, or some features can be ignored, or not implemented. In addition, the coupling, or direct coupling, or communication connection between the components shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be electrical, mechanical or other forms. of.

The unit described above as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit; it may be located in one place or distributed to multiple network units; Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present invention may all be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integration The unit can be implemented either in the form of hardware or in the form of hardware plus software functional units.

Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by program instructions related to hardware, the aforementioned program can be stored in a computer-readable storage medium, and when the program is executed, the execution includes: The steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: a removable storage device, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, etc. A medium on which program code is stored.

Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention may be embodied in the form of software products in essence or the parts that make contributions to the prior art. The computer software products are stored in a storage medium and include several instructions for A computer device (which may be a personal computer, a server, or a network device, etc.) is caused to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: a removable storage device, a ROM, a RAM, a magnetic disk or an optical disk and other mediums that can store program codes.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention. should be included within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims (10)

1. A dynamic resource allocation method of a space-based cloud computing architecture is characterized by comprising the following steps:

sensing the state parameters of the current network by using the existing network resources and the historical network data to generate a network state sensing result;

according to the network state perception result, detecting the traffic condition of each node and each link, the user access amount and the consumption of each application on the traffic under the current network topology structure, and carrying out classification and identification on the traffic to generate a traffic classification and identification result;

summarizing and inducing the flow characteristics of each node and each link under the current network topology structure according to the flow classification recognition result, predicting the flow condition based on the geographic position, the time distribution and the access intensity degree, and generating a flow prediction result;

and according to the flow prediction result, generating a resource configuration scheme through visualization of a big data analysis platform.

2. The method of claim 1, wherein the sensing of the state parameters of the current network using existing network resources and network history data further comprises:

and sensing the state parameters of the current network by an in-band network telemetry technology, wherein the state parameters are one or more of network physical topology, queue capacity and single-hop time delay.

3. The method of claim 1, wherein the detecting traffic conditions of each node and link, the user access amount, and the consumption amount of each application on the traffic under the current network topology, and classifying and identifying the traffic further comprises:

the method for indexing the data stream by the Hash function is adopted to detect the passing elephant stream on line at high speed, when the data stream arrives, the Hash index is utilized to carry out fast statistics on the scale of the data stream, and the data stream with the scale exceeding a certain threshold value is judged to be the elephant stream; and/or

And classifying and identifying the flow through a machine learning method based on the flow characteristics.

4. The method of claim 3, wherein the classification and identification of the traffic is performed by a machine learning method based on the traffic characteristics, and further comprising:

defining characteristics by which traffic can be identified and differentiated;

training an ML classifier which can associate the feature set of the flow with known classes, applying an ML algorithm and classifying unknown flow according to a rule model which is trained well in the prior learning.

5. The method for dynamically allocating resources in a space-based cloud computing architecture according to claim 1, wherein traffic characteristics of each node and each link under a current network topology are summarized and summarized according to traffic classification recognition results, and traffic conditions based on geographical location, time distribution, and access density are predicted, further comprising:

and predicting the traffic condition based on the geographic position, the time distribution and the access intensity degree by a regression learning method and/or a reinforcement learning method.

6. The method for dynamically configuring resources of the space-based cloud computing architecture according to claim 1, wherein the resources include computing resources and storage resources, the computing resources are one or more of a CPU, a GPU and an FPGA, and the storage resources are one or more of a mechanical hard disk, a solid state hard disk, a liquid state hard disk and an optical hard disk.

7. A dynamic resource allocation system of a space-based cloud computing architecture, the dynamic resource allocation system of the space-based cloud computing architecture comprising: the dynamic resource allocation method program of the space-based cloud computing architecture is executed by the processor to realize the following steps:

sensing the state parameters of the current network by using the existing network resources and the historical network data to generate a network state sensing result;

according to the network state perception result, detecting the traffic condition of each node and each link, the user access amount and the consumption of each application on the traffic under the current network topology structure, and carrying out classification and identification on the traffic to generate a traffic classification and identification result;

summarizing and inducing the flow characteristics of each node and each link under the current network topology structure according to the flow classification recognition result, predicting the flow condition based on the geographic position, the time distribution and the access intensity degree, and generating a flow prediction result;

and according to the flow prediction result, generating a resource configuration scheme through visualization of a big data analysis platform.

8. The system of claim 7, wherein the sensing of the state parameters of the current network using existing network resources and network history data further comprises:

and sensing the state parameters of the current network by an in-band network telemetry technology, wherein the state parameters are one or more of network physical topology, queue capacity and single-hop time delay.

9. The system for dynamically configuring resources in space-based cloud computing architecture according to claim 7, wherein the traffic condition of each node and link, the user access amount, and the consumption amount of each application on the traffic under the current network topology are detected and classified, and the system further comprises:

the method for indexing the data stream by the Hash function is adopted to detect the passing elephant stream on line at high speed, when the data stream arrives, the Hash index is utilized to carry out fast statistics on the scale of the data stream, and the data stream with the scale exceeding a certain threshold value is judged to be the elephant stream; and/or

And classifying and identifying the flow through a machine learning method based on the flow characteristics.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium includes a program of a dynamic resource allocation method for space-based cloud computing architecture, and when the program of the dynamic resource allocation method for space-based cloud computing architecture is executed by a processor, the steps of the method for dynamic resource allocation for space-based cloud computing architecture as claimed in any one of claims 1 to 7 are implemented.

Images