WO2023179180A1 - 一种网络虚拟化体系结构以及虚拟化方法 - Google Patents

一种网络虚拟化体系结构以及虚拟化方法 Download PDF

Info

Publication number
WO2023179180A1
WO2023179180A1 PCT/CN2023/070889 CN2023070889W WO2023179180A1 WO 2023179180 A1 WO2023179180 A1 WO 2023179180A1 CN 2023070889 W CN2023070889 W CN 2023070889W WO 2023179180 A1 WO2023179180 A1 WO 2023179180A1
Authority
WO
WIPO (PCT)
Prior art keywords
space
virtualization
resources
virtual
technology
Prior art date
Application number
PCT/CN2023/070889
Other languages
English (en)
French (fr)
Inventor
匡立伟
尹山
徐安然
李文超
Original Assignee
烽火通信科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 烽火通信科技股份有限公司 filed Critical 烽火通信科技股份有限公司
Publication of WO2023179180A1 publication Critical patent/WO2023179180A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/04—Network management architectures or arrangements
    • H04L41/042—Network management architectures or arrangements comprising distributed management centres cooperatively managing the network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/04—Network management architectures or arrangements
    • H04L41/044—Network management architectures or arrangements comprising hierarchical management structures
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0803—Configuration setting
    • H04L41/0823—Configuration setting characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present invention relates to the field of communication technology, and in particular to a network virtualization architecture and a virtualization method.
  • Information technology includes two larger technology systems, one is the information processing technology system represented by cloud computing, and the other is the information transmission technology system represented by the network.
  • Information processing technology focuses on the research of efficient algorithms to achieve large-scale high-performance data processing and mine valuable information from data. This type of technology includes distributed computing, parallel algorithms, data mining, knowledge discovery, etc.
  • Information transmission technology focuses on the research of connection protocols and high-performance devices to support higher speed, larger capacity, and longer distance data packet transmission. Such technologies include network protocols, digital signal processing, optical amplification technology, and error correction coding technology.
  • the first type is to make computing storage resource requests to the cloud computing center
  • the second type is to make transmission resource requests to the communication network.
  • This shallow-level cloud-network collaboration technology has two obvious shortcomings. First, users need to coordinate the technical details of two types of resources at the same time, making it difficult for users to focus on business processes. Secondly, whether it is "the network better cooperates with the cloud” or “the cloud better cooperates with the network”, it is difficult to achieve global overall scheduling of cloud network resources. Computing, storage and transmission resources achieve optimal utilization locally, but still cannot guarantee the global optimal utilization. excellent.
  • the present invention provides a network virtualization architecture and virtualization method, comprehensively analyzes and summarizes cloud computing center and communication network related technologies, and builds a new system based on network virtualization technology.
  • the structure uniformly describes computing, storage, and transmission resources, and supports the integration of resources in physical space, virtual space, and application space.
  • the network virtualization architecture proposed by the present invention not only conforms to the current technical trend of smooth evolution of cloud network architecture, but also supports the deep integration of computing, storage, and transmission resources. At the same time, it supports users to simultaneously apply for required resources in a unified quantified space, and supports the global optimization of resources. Excellent overall scheduling.
  • the present invention provides a network virtualization architecture, including physical space, virtual space and application space, wherein:
  • the physical space includes computing resources, storage resources, and transmission resources, and realizes the integration of computing, storage, and transmission resources.
  • the physical space also reports device resource status data to the virtual space;
  • the virtual space implements the virtualization of various resources and describes them in the virtualization model to present them to the application space; the virtual space also issues demand instructions to specific users after receiving the requirements issued by the application space. Execute on physical device or virtual device;
  • the application space obtains the virtualization model described by the virtual space and provides it to various business scenarios; the application space also accepts requirements, integrates the requirements and delivers them to the virtual space.
  • the virtual space includes an orchestrator, a controller, and virtual computing resources, virtual storage resources, and virtual transmission resources corresponding to the physical space;
  • the application space includes user applications, computing power requirements, storage capacity requirements, Transmission capability requirements;
  • the interface between the physical space and the virtual space includes a device status interface and a management and control instruction interface;
  • the interface between the application space and the virtual space includes a cloud network capability interface and a user demand interface.
  • the physical space, virtual space, application space, device status interface, management and control instruction interface, cloud network capability interface, and user demand interface form a closed loop. Specifically:
  • the application space realizes the integration of computing power demand, storage capacity demand and transmission capacity demand. After the user puts forward the resource demand, the application space will match it, find the corresponding virtual resource, and send it to the virtual space through the user demand interface;
  • the virtual space realizes the integration of virtual computing resources, virtual storage resources, and virtual transmission resources. After receiving user requirements, it is processed through the orchestrator and controller to form specific operation instructions, which are issued to specific physical devices or devices through the management and control instruction interface. Execute on virtual device;
  • the physical space realizes the integration of three types of entity resources: computing resources, storage resources, and transmission resources. Part of the status data in the business disk of the physical space is processed on the computing storage disk, and the other part is reported to the virtual space through the device status interface;
  • the virtual space can grasp the performance status of computing resources, storage resources, and transmission resources in the physical space in real time.
  • the computing, storage, and transmission resource capabilities of the virtual space are uniformly described in the virtualization model, and through the cloud network
  • the capability interface is presented to the application space, and the application space provides it to various business scenarios.
  • the process of realizing automatic integration and scheduling of computing, storage, and transmission resources in the physical space, the virtual space, and the user space includes:
  • the virtual space orchestrator and controller After completing the unified description of resource quantification indicators, the virtual space orchestrator and controller convert the resource quantification indicators into specific operation instructions;
  • Computing, storage, and transmission equipment receive and execute various operating instructions.
  • computing resources storage resources, and transmission resources, including:
  • One or more computing storage disks are arranged in a physical space, and the physical space also includes a main control disk and several business disks;
  • Install and deploy basic software including one or more of the operating system, data tool software, digital twin modeling framework, intelligent training inference framework, and algorithm analysis and processing framework; among them, the data tool software calls the functions provided by the operating system to access the central One or more of processors, memory, disks, and peripheral devices; the digital twin modeling framework, the intelligent training inference framework, and the algorithm analysis and processing framework call data tool software to obtain various types of data;
  • the business disk is used to collect device status data. If the collected data is related to this network element, it is uploaded to the computing storage disk for edge processing; if it is related to other network elements, it is uploaded to the main control disk and handed over by the main control disk. to the management and control platform; after the computing storage disk receives the device status data, it is deployed according to the data processing needs. If the simulation modeling operation is performed, the data is forwarded to the digital twin modeling framework; if edge artificial intelligence inference is performed, the data is forwarded. Give the intelligent training inference framework; if alarm correlation analysis or performance degradation analysis is performed, the data will be forwarded to the algorithm analysis and processing framework.
  • the virtualization model includes a three-layer, three-sided network virtualization model and a seventh-order tensor model, where:
  • the three-layer and three-sided network virtualization model includes three types of resources, three layers of space, and three technologies.
  • the three types of resources include computing resources, storage resources, and transmission resources.
  • the three layers of space include physical space, virtual space, and application. space, the three technologies include virtualization technology, intelligent operation and maintenance technology, and endogenous security technology; the three types of resources are located within the three-layer space, and the three technologies act on the three-layer space and all Describe the three types of resources;
  • the seventh-order tensor model provides a unified description and quantitative representation of three types of resources located in the three-layer space.
  • the three-type technologies are used to process various elements in the seventh-order tensor model, including: based on virtual Virtualization technology realizes the virtualization of physical space computing resources, storage resources, and transmission resources, realizes cross-domain coordination and scheduling of virtualized resources based on intelligent operation and maintenance technology, and builds a safe and reliable virtualization system based on endogenous security technology.
  • the seventh-order tensor model is described as T ⁇ R I1 ⁇ I2 ⁇ I3 ⁇ I4 ⁇ I5 ⁇ I6 ⁇ I7 , where R represents the real number field, and I1, I2, I3, I4, I5, I6, and I7 represent the tensor
  • the seven levels of the quantitative model represent time, physical space, virtual space, application space, computing resources, storage resources, and transmission resources respectively.
  • the present invention also provides a virtualization method that uses virtualization technology to virtualize various resources, wherein the virtualization technology includes functional fidelity technology and functional simulation technology, and the functional fidelity technology Including scheduling fidelity technology and virtual fidelity technology.
  • the functional simulation technology includes function mapping simulation technology and function fitting simulation technology. For a specific physical entity resource, when selecting virtualization technology, the scheduling fidelity technology and virtual fidelity technology should be followed. Select the real technology, function mapping simulation technology, and function fitting simulation technology in the order of priority.
  • the selection of virtualization technology also includes three transition modes: P transition, F transition, and FP transition. Specifically:
  • call the virtualization model pass in real-time data, create virtualization instances, and build a virtual network
  • the application space After the application space completes the specific tasks of various application scenarios, it provides feedback on the virtualization effect and optimizes the virtualization model based on the virtualization effect.
  • the present invention proposes a new network virtualization architecture to realize cross-domain unified scheduling and control of computing resources, storage resources, and transmission resources, support end-to-end global orchestration of services, and support intelligent operation and maintenance of resources.
  • the overall network virtualization architecture realizes the deep integration of cloud computing environment and communication network.
  • the present invention adopts a new tensor-based three-layer, three-sided network virtualization model and a seventh-order tensor model to uniformly describe computing, storage, and transmission resources in a high-order multi-dimensional tensor space, and clearly define physical space, virtual space,
  • virtualization technology, intelligent operation and maintenance technology, and endogenous security technology are introduced to solve the problems of current network architecture's poor ability to allocate resources on demand, difficulty in cross-domain collaboration of resources, and time-consuming and inefficient resource operation and maintenance.
  • the present invention proposes two major categories and four sub-categories of network virtualization technologies.
  • a high-precision virtualization model can be constructed by selecting appropriate virtualization technologies.
  • virtualized networks are created and business verification or resource operations are performed, which can realize automatic processing of network services, automatic maintenance of network resources, and automatic optimization of network performance.
  • the present invention also adopts a virtualization instance automatic optimization mechanism to optimize the virtualization instance through cross-layer collaborative feedback technology of physical space, virtual space, and application space to improve the simulation degree of the virtualization model to the physical entity.
  • the computing, storage, and transmission resource fusion method proposed by the present invention can achieve deep resource fusion in physical space, virtual space, and application space respectively.
  • the physical space supports the distribution of data between a single disk for computing storage and a single disk for transmission services.
  • the virtual space deploys business orchestrators and resource controllers, supports global business orchestration and cross-domain resource control, supports the creation, combination, and optimization of virtualized instances, and realizes the on-demand opening of network capabilities through virtualized instances.
  • users can make concise and precise requests for computing, storage and transmission resources without knowing too much technical background, allowing users to pay more attention to the business, improving user perception, and ensuring the accuracy and consistency of resource requests.
  • Figure 1 is a schematic diagram of a network virtualization architecture oriented to the integration of computing, storage and transmission resources provided in Embodiment 1 of the present invention
  • Figure 2 is a schematic diagram of a network virtualization technology tree provided by Embodiment 3 of the present invention.
  • Figure 3 is a schematic diagram of the optimization of physical entity virtualization modeling and virtualization examples provided in Embodiment 4 of the present invention.
  • Figure 4 is a schematic diagram of the integration of computing, storage and transmission resources in the physical virtual application space provided by Embodiment 5 of the present invention.
  • Figure 5 is a schematic diagram of automatic resource integration and scheduling based on virtualization technology provided in Embodiment 6 of the present invention.
  • Figure 6 is a schematic diagram of the integration of computing, storage and transmission resources in the physical space provided by Embodiment 7 of the present invention.
  • the present invention is an architecture of a specific functional system. Therefore, in the specific embodiments, the functional logical relationship of each structural module is mainly explained, and the specific software and hardware implementation methods are not limited.
  • Cloud computing resources include computing resources and storage resources, such as central processing units and artificial intelligence chips, etc. are all computing resources.
  • Storage arrays, hard disks, memory, etc. are all storage resources.
  • Network resources mainly refer to transmission resources, including optical fibers, network cables, single disks of communication equipment, optical modules, optical devices, network protocols, etc.
  • the core of “cloud network convergence” is to achieve the integration of computing, storage, and transmission resources, and the goal of “cloud network convergence” is to provide users with computing, storage, and transmission resources efficiently, flexibly, intelligently, and securely.
  • embodiments of the present invention propose a new network virtualization architecture to realize computing resources , deep integration and collaborative scheduling of storage resources and transmission resources, supporting the integration and scheduling of the above three types of resources in physical space, virtual space, and application space, achieving resource efficiency and flexibility based on virtualization technology, intelligent operation and maintenance, and endogenous security technology , intelligent and safe supply goals.
  • Embodiment 1 of the present invention provides a network virtualization architecture oriented to the integration of computing, storage, and transmission resources, which includes a physical space, a virtual space, and an application space, wherein: the physical space realizes the integration of computing, storage, and transmission resources, and combines device resources with Status data is reported to the virtual space; the virtual space implements virtualization of various resources and describes them in the virtualization model to present them to the application space; the virtual space also receives requests from the application space Then the demand instructions are issued to specific physical devices or virtual devices for execution; the application space obtains the virtualization model described in the virtual space and provides it to various business scenarios; the application space also accepts the requirements, integrates the requirements and downloads them. Sent to virtual space.
  • This preferred embodiment proposes two types of virtualization technologies, namely function-fidelity virtualization technology (function-fidelity technology) and function-simulation virtualization technology (function-simulation technology), both of which are used to realize various resources in the virtual space.
  • the virtualization technology used in virtualization The input and output of the virtualization model built based on functional fidelity technology are consistent with the physical entity model, and the internal processing mechanism is consistent with the physical entity. It only simulates hardware operations through software methods. The input and output of the virtualization model built based on functional simulation technology are not completely consistent with the physical entity model. It mainly simulates the operation mode of the physical entity through software to verify, monitor, and optimize the function and performance of the physical entity.
  • the functional fidelity virtualization technology proposed in this embodiment is further subdivided into two specific technologies, namely scheduling fidelity technology and virtual fidelity technology.
  • Scheduling fidelity technology virtualizes multiple logical network slices in the physical network, and each slice meets specific types of business requirements.
  • Virtual fidelity technology virtualizes network element functions based on dedicated hardware and then deploys them on general hardware to form a virtual network element pool.
  • the function simulation virtualization technology proposed in this embodiment is also subdivided into two specific technologies, namely function mapping simulation technology and function fitting simulation technology.
  • functional mapping simulation technology translates and maps the communication mechanism of network entities in physical space, it builds a corresponding digital twin model in virtual space.
  • Functional fitting simulation technology mainly focuses on network functions that are difficult to accurately describe the communication mechanism. It performs fitting based on input and output data to build a digital twin model.
  • virtualization resources can be created by selecting from the above virtualization technologies based on the characteristics of the physical entity.
  • the virtualization model in this embodiment includes a three-layer, three-sided network virtualization model and a seventh-order tensor model.
  • the tensor-based three-layer and three-plane network virtualization model (shown on the left side of Figure 1) is called Tensor based Three Layer and Three Plane Model in English, and the English abbreviation is TL 3 P 3 M.
  • This three-layer and three-plane network virtualization model includes three types of resources, three levels of space, and three types of technologies.
  • the three types of resources are located within the three levels of space.
  • the three technologies act on the three levels of space and the three types of resources at the same time, where , the three types of resources include computing resources, storage resources, and transmission resources.
  • the three-layer space includes physical space, virtual space, and application space.
  • the three technologies include virtualization technology, intelligent operation and maintenance technology, and endogenous security technology.
  • embodiments of the present invention realize virtualization of physical space computing resources, storage resources, and transmission resources based on virtualization technology, realize cross-domain coordination and scheduling of virtualized resources based on intelligent operation and maintenance technology, and build a stable and reliable virtualization system based on endogenous security technology. .
  • the network virtualization architecture of this embodiment also includes a seventh-order tensor model corresponding to the three-layer three-plane network virtualization model (as shown on the right side of Figure 1).
  • Class resources are uniformly described and quantified, and the three types of technologies are used to process various elements in the seventh-order tensor model.
  • the seventh-order tensor model is described as T ⁇ R I1 ⁇ I2 ⁇ I3 ⁇ I4 ⁇ I5 ⁇ I6 ⁇ I7 , where R represents the real number domain, I1, I2, I3, I4, I5, I6, I7 Represents the seven orders of the tensor model, which respectively represent time, physical space, virtual space, application space, computing resources, storage resources, and transmission resources.
  • This preferred embodiment also proposes three resource fusion methods, including a physical space resource fusion method, a virtual space resource fusion method, and an application space resource fusion method.
  • Traditional communication equipment contains many business disks and only has a single transmission function.
  • This preferred embodiment proposes a physical space resource fusion method.
  • a single computing and storage disk is added to the communication device. Part of the single business disk data is uploaded to the single computing and storage disk, processed directly on the device side, and part of it is uploaded to the management and control platform.
  • the virtual space resource fusion method proposed in this preferred embodiment realizes cross-domain overall scheduling of virtual computing resources, virtual storage resources, and virtual transmission resources, and can optimally schedule various virtual resources according to templates issued by the global service orchestrator.
  • the application space resource fusion method proposed in this preferred embodiment allows industry users to directly request computing, storage and transmission resources based on business needs without having the technical background of underlying resource capabilities.
  • the network virtualization architecture as a whole converts user needs into Resource requirements are simulated and verified based on the virtualized resource pool for scheduling and allocation. After accuracy, the corresponding physical physical devices are issued through the controller to realize the allocation of computing, storage and transmission resources.
  • communication equipment manufacturers develop and produce virtualization models, including virtualized computing resource models, virtualized storage resource models, and virtualized transmission resource models, while developing and producing physical physical equipment.
  • virtualization models are managed and controlled uniformly by the network virtualization architecture.
  • the data collected and reported by the physical entity in real time is integrated with the virtualization model to form a series of virtualization instances.
  • the virtualization instances in the virtual space are
  • the physical devices in the physical space correspond to each other, supporting real-time presentation, historical retrospection, and forward-looking prediction of the operating mechanism and health status of the physical physical devices.
  • the communication network management and control platform creates a performance optimization plan based on intelligent technology, and creates a virtualization instance in the virtual space to verify the performance optimization plan. If the optimization can be achieved The effect is that the operation instructions can be directly sent to the physical physical device.
  • many operations in communication networks take a long time, but once a network failure or performance degradation occurs, the network needs to be able to recover quickly.
  • the virtualization instance accurately matches the existing network health status. Potential network hazards or deterioration can be predicted based on the virtualization instance, and corresponding processing plans can be formed and verified.
  • this Embodiment 2 further details the three-layer, three-sided network virtualization model and seventh-order tensor model of the network virtualization architecture. Specific instructions.
  • the entire model is presented as a cube.
  • the bottom layer is the physical space where physical equipment, including computing, is deployed.
  • Equipment such as servers
  • storage devices such as disk arrays
  • transmission equipment such as optical transmission equipment, data communication equipment.
  • the middle layer is the virtual space, which deploys business orchestrators, cross-domain controllers, single-domain controllers, virtual computing resources, virtual storage resources, and virtual transmission resources created based on virtualization technology.
  • the top layer is the application space, which contains a series of user applications, such as distributed machine learning.
  • the right side of the cube includes three faces, which correspond to three technologies.
  • Virtualization technology simulates and models the operating mechanisms of various physical devices in the physical space, builds corresponding virtual resources, and deploys them in the virtual space.
  • Virtualization technology models various business scenarios in the application space, abstracts the main characteristics of various applications, and recommends efficient resource allocation models to users based on the main characteristics.
  • Intelligent operation and maintenance technology can realize automated processing of various resource planning stages, activation stages, and maintenance stages in the time dimension.
  • the network virtualization architecture ensures the security of various physical and virtual resources and business scenarios in physical space, virtual space, and application space.
  • the right side of Figure 1 shows the seventh-order tensor network virtualization architecture model (seventh-order tensor model) proposed by the embodiment of the present invention.
  • This model can quantitatively represent the various elements in the three-layer, three-sided network virtualization model and establish various relationship between elements.
  • three types of resources i.e., computing resources, storage resources, and transmission resources
  • spaces i.e., physical space, virtual space, and user space
  • Three types of technologies i.e., virtualization technology, intelligent operation and maintenance technology, and endogenous security technology
  • process elements in the tensor model to achieve efficient, flexible, intelligent, and secure resource supply.
  • the seventh-order tensor model in this embodiment is described as T ⁇ R I1 ⁇ I2 ⁇ I3 ⁇ I4 ⁇ I5 ⁇ I6 ⁇ I7 , where R represents the real number domain, I1, I2, I3, I4, I5, I6, I7 Represents the seven orders of the tensor model, which respectively represent time, physical space, virtual space, application space, computing resources, storage resources, and transmission resources.
  • Each order of the tensor contains many dimensions.
  • the seven dimensions of the seventh order can determine the specific position of the object in the seventh-order tensor model.
  • the tensor element value at this position is a real number, and the real value corresponds to the specific characteristic value of the object itself.
  • Table 1 below uses machine learning application scenarios, controller systems, and communication equipment computing disks as examples to explain how to obtain values for the tensor elements of the seventh-order tensor model.
  • Table 1 Examples of numerical values for tensor elements of the seventh-order tensor model
  • the second to fourth rows represent three objects
  • the third to ninth columns represent the seven orders of the tensor model
  • the tenth column is the specific numerical value of the object.
  • the seventh-order tensor model proposed in this embodiment is a general model, and specific indicators representing the values of each dimension of the seventh-order tensor can be defined according to actual application requirements. For example, if the current communication network physical equipment adopts a minute-level data reporting mode and reports full performance data every fifteen minutes, you can specify that each integer of the first-order dimension value of the tensor model represents a period of fifteen minutes.
  • the three-dimensional index values of physical space, virtual space, and application space can be defined according to specific needs.
  • dimension 1 of the application space I4 is defined as the machine learning business scenario
  • dimension 2 is defined as the virtual reality application scenario, so the value in row 2, column 6 of Table 1 is 1.
  • the controller is located in the virtual space, and the tensor I3 with order dimension 1 represents the controller object.
  • the value 1 in the 4th row and 4th column of Table 1 indicates that the I2 order of the tensor model corresponds to the calculated single disk object.
  • the I5 order of the tensor model indicates the number of CPU cores of the computing resource.
  • the value 8 in row 2, column 7 of Table 1 indicates that the computing resource required for machine learning is an 8-core CPU.
  • the value 2 in row 2, column 8 of Table 1 indicates that the storage resources required for machine learning are 2T hard disk space
  • the value 4 in row 2 indicates that the transmission resources required for machine learning are 2 network slices.
  • Row 2 and column 10 of Table 1 correspond to the tensor element value.
  • the value 1 represents a machine learning application.
  • the object in row 3 of Table 1 is the controller, and the tensor element value 2 indicates that two sets of controllers are deployed in the virtual space. During the specific implementation process, two sets of controllers are deployed in active and backup mode to prevent network business interruption due to controller failure.
  • the resources and operating status configured by the two sets of controllers are consistent, so the two sets of controllers are controlled in Tensor
  • the dimensions corresponding to each level of the model are also the same.
  • the definitions of the controller and other stages of the single computing disk are the same as those of machine learning, so we will not go into details here.
  • Table 1 of this embodiment describes by way of example how to map the requirements of various objects to various dimensions of the tensor model during the implementation process.
  • T represents the tensor
  • the seven values in parentheses respectively represent the values of each dimension of the seven tensor orders.
  • the right side of the equation is the tensor element value.
  • the fifth element in the brackets of T(2,0,0,1,8,2,4) represents the fifth order I5 dimension of the tensor, which actually means that the computing resource is an 8-core central processing unit CPU (Central Processing Unit).
  • CPU Central Processing Unit
  • machine learning and controllers are the demanders of computing, storage and transmission resources. Machine learning objects need to consume resources to implement artificial intelligence learning and training, and controllers consume resources to perform various tasks such as communication network path calculations.
  • the single computing disk in Table 1 is the resource provider and can provide computing resources and storage resources.
  • the seventh-order tensor model proposed in this embodiment can provide a specific quantitative description of the computing, storage, and transmission resources required by various objects in the physical space, virtual space, and application space, and can also describe the resources that various objects can provide. Capability values are quantitatively described, thereby achieving a unified quantitative description of computing, storage, and transmission resource requirements and supply capabilities in a tensor model.
  • the greatest beneficial effect of constructing the tensor model proposed in this embodiment is to lay the foundation for the subsequent establishment of complex quantitative relationships among the three types of resources: computing, storage, and transmission.
  • computing, storage, and transmission only by quantitatively describing computing, storage, and transmission resources and achieving unified representation in a space can it be possible to establish quantitative relationships between various resources, and establishing quantitative relationships is the core of achieving optimal resource scheduling.
  • virtualization technology abstractly models computing, storage, and transmission resources and expresses them in an orderly manner in a unified space in a quantified form; artificial intelligence technology establishes a network of computing, storage, and transmission resources based on tensor space elements.
  • the quantitative relationship between cloud computing centers and communication networks enables intelligent operation and maintenance of the entire life cycle of "planning, construction, maintenance, optimization, and operation"; endogenous security technology establishes the relationship between computing, storage, and transmission resources based on tensor space. Quantity relationship, clearly assess the impact that various security risks may have on user business applications, formulate relevant security plans, and activate security plans once security problems occur to ensure that users can still use computing, storage, and transmission resources.
  • Embodiment 3 Based on the network virtualization architecture provided in Embodiment 1 for the integration of computing, storage and transmission resources, this Embodiment 3 provides a more detailed description of the virtualization technology adopted.
  • FIG. 2 it is a network virtualization technology tree proposed in this embodiment.
  • This technology tree includes a root, two branches, and four leaves.
  • the root of the tree represents network virtualization technology
  • the two branches represent functional fidelity technology and functional emulation technology.
  • the four leaves are scheduling fidelity technology, virtual fidelity technology, function mapping simulation technology, and function fitting simulation technology.
  • this embodiment proposes the method of "fidelity first, simulation second".
  • Select virtualization technology from the four leaves in order from left to right, that is, select in the order of 1234 (scheduling fidelity technology, virtual fidelity technology, function mapping simulation technology, and function fitting simulation technology).
  • For a specific physical entity resource first choose scheduling fidelity technology for virtualization. If it is difficult to implement, then choose virtual fidelity technology for abstract modeling. If the fidelity technology cannot be applied to the physical entity resource, then the function mapping technology is selected to build the virtual entity. If it still cannot be implemented, the function fitting technology is selected to build the virtual entity.
  • the virtualization technology represented by the four leaves in Figure 2 can jump in the order from right to left after certain conditions are met, that is, in the order of 4321.
  • This embodiment proposes three transition modes, namely P transition, F transition, and FP transition.
  • the network virtualization technology tree proposed in this embodiment includes four types of virtualization technologies, one virtualization technology selection method, and three transition methods. One virtualization technology selection method and three transition methods were described above. The following describes the four types of virtualization technologies in detail.
  • the communication equipment entities in the physical space are constructed into corresponding virtual entities through virtualization technology and deployed in the virtual space. If the functions of the virtual entity correspond to the functions of the physical device entity, functional fidelity technology needs to be adopted. If the virtual entity does not need to completely map the operating mechanism of the physical device entity, functional simulation technology can be used.
  • the scheduling fidelity technology of this embodiment uses function mapping to collect and sort user requests for computing, storage, and transmission resources, formulate resource scheduling priorities, and allocate resources from high to low according to priority to meet user needs.
  • the lower left corner of Figure 2 is a diagram of scheduling fidelity technology. The two circles at the bottom represent physical resources, the top circle represents user resource requirements, and the middle box represents the scheduler. The scheduler understands the availability of all physical resources, schedules them based on the resource requirements submitted by users, and allocates appropriate resources to users. In the lower left corner of Figure 2, the scheduler allocates the physical resources represented by the circle on the left to complete the user's needs.
  • Table 2 takes the delay path resource requirements as an example to illustrate the scheduling fidelity virtual technology. If the user needs to complete an operation within 15 milliseconds, the calculation time is 5 milliseconds, and the data transmission time between the sender and the receiver needs to be less than 5 milliseconds. Based on the three-layer, three-sided network virtualization model and seventh-order tensor model constructed in this embodiment, the physical entities on all transmission paths of the sender and the receiver are searched, and the delay of data processing by the physical entities on each path is accumulated to obtain this Total path delay. For example, a path connected through 400 kilometers of optical fiber has a transmission delay of 2 milliseconds. Involving two optical amplifiers, the transmission delay is 0.2 microseconds.
  • the transmission delay of this path is about 5 milliseconds.
  • the scheduler can allocate this path and the corresponding sender-receiver to the user to meet the user's needs for calculation delay and transmission delay.
  • the second picture on the lower left side of Figure 2 describes virtual fidelity technology.
  • the three circles on the bottom represent physical devices, and the two pentagons on the top represent virtual devices.
  • the physical device is built based on dedicated hardware to implement electrical and optical signal processing and data message transmission.
  • Virtual devices are based on general-purpose hardware, use clusters to improve processing performance, and simulate the data transmission functions of physical devices through software programs.
  • customer-side equipment CPE Customer Premise Equipment
  • vCPE virtualized Customer Premise Equipment
  • BRAS Broadband Remote Access Server
  • Virtual fidelity technology can map the operating mechanisms of physical devices and virtual devices to ensure that the functions of physical devices and virtual devices are consistent.
  • the virtual device built through virtual fidelity technology can be in the form of a box-type physical device and run virtual device software on a general-purpose central processor.
  • the virtual device software can also be deployed directly on the cloud data center virtual machine.
  • the specific form of the virtual device can be determined according to the user application scenario during the implementation process.
  • the third diagram below Figure 2 describes the functional mapping simulation technology. If the internal mechanism of physical space communication equipment can be understood in detail, the internal mechanism can be translated into a mathematical model through functional mapping simulation technology.
  • Table 3 uses an erbium-doped fiber amplifier as an example to illustrate the implementation steps of functional mapping simulation technology, including seven steps. The first step is to obtain the parameter values of the erbium-doped fiber amplifier, such as the gain spectrum and functional spectrum. The second step is to set the incident pump and signal light power. The third step is to calculate the fundamental mode distribution of the pump light and signal light. The fourth step is to calculate the incident pump and signal photon flux. There are formulas in the optical communication system theory. Relationship between photon flux and power.
  • the fifth step is to calculate the energy level distribution and calculate the lateral distribution of the gain coefficient.
  • the sixth step is loop iteration. Through instructions such as update and determination, the pump and signal light power distribution at the end of the amplifier fiber is calculated from the incident pump and signal light power. This step outputs the simulation model source code.
  • the seventh step is to deploy the simulation model on the virtualization platform, collect network data, perform data and model fusion operations, perform calculations through the simulation model, and output virtualization simulation results.
  • the last picture at the bottom of Figure 2 is a schematic diagram of functional fitting simulation virtualization technology. If it is impossible to understand the internal mechanism of physical space communication equipment in some cases, you can measure the input and output data of the communication equipment through instruments in the laboratory, build a model based on functional fitting simulation technology, and approximate the internal reality of the physical entity through fitting functions.
  • Function function The physical entity of the communication equipment in the figure contains five functional modules. After the first functional module is executed, it branches. The upper process requires two functional modules to process, and the lower process requires one functional module to process. After the two branches are processed, the final process A functional module summarizes the processing and outputs the results. Through function fitting simulation technology, the input data and output data are input and the simulation function is fitted.
  • the upper branch and two circle function modules are fitted into a triangular function module and a hexagonal function module, and the lower branch is fitted as A hexagonal functional module.
  • This figure is a schematic diagram used to vividly illustrate the process flow of functional fitting simulation technology. For example, if you perform virtual modeling of optical signal-to-noise ratio OSNR (Optical Signal Noise Ratio), you can use instruments to measure optical power, dispersion and other values in the laboratory as input, measure the optical signal-to-noise ratio value as output, and select Deep Neural The network DNN (Deep Neural Network) model is fitted to obtain the optical signal-to-noise ratio virtualization simulation model.
  • OSNR Optical Signal Noise Ratio
  • the network virtualization technology tree proposed in this embodiment includes a total of four types of virtualization technologies, a virtualization technology selection method, and three transition methods. Based on the above technologies and methods, the virtualization of physical entities can be most effectively realized, and the virtualization can be realized according to the situation.
  • the transformation updates the virtualized entity to achieve optimal performance and functionality, and lays the foundation for the subsequent integration of computing, storage and transmission resources.
  • this Embodiment 4 proposes a physical entity virtualization architecture.
  • An optimization method for models and virtualized instances which includes optimizing the built virtualized instances when performing virtualized modeling of physical entities.
  • the method in this embodiment includes four steps.
  • the first step is to clarify the physical entities deployed in the real network, and upload the physical entity's device status data, physical entity configuration data, etc. to the virtualization space (corresponding to "collection" in Figure 3).
  • the lower part of Figure 3 is an example of physical entities, including optical transmission equipment and optical fibers, various single disk models and parameters of optical transmission equipment, various module and device model parameters are uploaded to the virtual space, optical fiber type, optical fiber dispersion coefficient, optical fiber attenuation Coefficients and other parameters are uploaded to the virtual space.
  • the second step according to the relevant device status data and configuration data uploaded by the physical entity, sort out the operating mechanism of the entity, select the corresponding virtualization technology to build a virtualization model (build a virtualization model based on the network virtualization technology proposed in Embodiment 3, corresponding to Figure "Modeling” in 3).
  • a virtualization model based on the network virtualization technology proposed in Embodiment 3, corresponding to Figure "Modeling” in 3.
  • function mapping simulation technology can be used to build a virtualization model.
  • functional fitting simulation technology can be used to build a virtualized model.
  • the third step is to call the virtualization model according to the user space business scenario requirements, pass in real-time data, create a virtualization instance, and build a virtual network (corresponding to "call" in Figure 3).
  • Figure 3 shows an example of a virtualized network, including six network nodes. This virtualized network can support application scenarios such as distributed machine learning.
  • the application space After the application space has completed the specific tasks of various application scenarios, it provides feedback on the virtualization effect and optimizes the virtualization model according to the virtualization effect (corresponding to "virtualization effect feedback", “adjustment combination strategy”, “Optimize virtual algorithm”, “Update acquisition mode”).
  • the optimization of virtualization models is divided into three types of situations.
  • the first type of situation updates the data collection mode, increases the frequency of data collection and reporting, or adds new data indicators to improve the model virtualization effect.
  • the second type of situation optimizes the virtual algorithm, such as enhancing or improving the function mapping method and adopting better fitting functions, so as to achieve the goal of improving simulation accuracy.
  • the third type of situation adjusts the combination strategy. This type of situation is used for complex business scenario simulation, and the model virtualization effect is improved by calling different virtualization models, such as increasing virtual computing resources while reducing virtual storage resources.
  • the physical entity virtualization modeling and virtualization instance upgrade and optimization method proposed in this embodiment is a dynamic and automatic method.
  • the traditional method is static and manual. After building the virtualization model, it passively waits for calls, and then passively waits for the next round of adjustments, and the adjustment strategy is relatively simple.
  • virtualization instances are generated in the virtual space, and all instances provide computing, storage, and transmission capabilities to users after being integrated in the virtualization space. After these capabilities are provided to users, they will be actively upgraded and optimized based on user feedback.
  • the optimization methods include updating the data collection mode, optimizing the virtual algorithm, and adjusting the combination strategy proposed in this embodiment.
  • the optimized virtual algorithm may adopt the virtualization technology transition method in Embodiment 3 of the present invention during implementation.
  • the dynamic automatic upgrade optimization method in this embodiment can improve the virtualization model in real time, create high-precision virtualization instances in real time, and ensure that the virtual computing, virtual storage, and virtual transmission resources created in the virtual space can meet the diverse services of users in real time. Application requirements to ensure cloud network integration service quality.
  • this Embodiment 5 proposes three integration methods of computing, storage, and transmission resources to implement resource integration in physical space, virtual space, and application space respectively.
  • Figure 4 shows a schematic diagram of the integration of computing, storage and transmission resources in the physical and virtual application space in this embodiment.
  • the bottom part of Figure 4 is the physical space P (Physical Space), which deploys physical equipment for computing, storage, and transmission resources. That is, the physical space includes computing resources, storage resources, and transmission resources.
  • the physical space computing storage transmission resource integration method proposed in this embodiment refers to adding computing and storage unit disks (which can be referred to as computing storage disks) on the transmission communication equipment. Part of the status data on the business disk is uploaded to the computing storage disk, and part of it is uploaded to the management and control disk. platform. If the status data on the business disk is only related to the communication device, upload the computing storage disk and process it on the device side. If the status data on the service disk is related to other network elements, it is uploaded to the management and control platform for centralized processing.
  • the computing storage disk deploys the artificial intelligence learning and training framework to support lightweight artificial intelligence model training and inference.
  • the middle of Figure 4 is the virtual space V (Virtua Space), where the orchestrator and controller are deployed.
  • virtual computing resources, virtual storage resources, and virtual transmission resources are constructed based on the network virtualization technology proposed in Embodiment 3. These three types of virtual resources are Also deployed in virtual space. Virtual resources correspond to physical resources.
  • virtual resources constructed based on scheduling fidelity technology, function mapping simulation technology, and function fitting simulation technology ultimately execute actual instructions in the corresponding of physical resources.
  • Virtual resources built on virtual fidelity technology can execute actual instructions.
  • there are two types of interfaces between virtual space and physical space The device status is uploaded from the physical space to the virtual space through the device status interface S-flow (State flow). Management and control instructions are issued from the virtual space to the physical space through the management and control instruction interface M-flow (Management and control flow).
  • the upper part of Figure 4 is application space A (Application space), which includes various user applications, such as distributed machine learning applications and cloud-based virtual reality applications.
  • the application space stores users' resource requirements, including computing power requirements, storage capacity requirements, and transmission capacity requirements.
  • the computing, storage and transmission resource capabilities are uniformly described in the three-layer, three-sided network virtualization model and the seventh-order tensor model. After the user submits the computing, storage and transmission capability requirements, the application space will search the tensor model and match the required resources and allocate them to users.
  • the interfaces between application space and virtual space include cloud network capability interface C-flow (Capability flow) and user demand interface R-flow (Requirement flow).
  • three spaces physical space, virtual space, and application space
  • four types of interfaces device status interface, management and control command interface, cloud network capability interface, and user demand interface
  • the application space realizes the integration of computing power requirements, storage capacity requirements, and transmission capacity requirements. After the user puts forward resource requirements, the application space will match them, find the corresponding virtual resources, and send them to the virtual space through the user demand interface R-flow.
  • the virtual space realizes the integration of virtual computing resources, virtual storage resources, and virtual transmission resources. After receiving user needs, it is processed through the orchestrator and controller to form specific operation instructions, which are issued to specific physical devices or devices through the management and control interface M-flow. Execute on virtual device.
  • the physical space realizes the integration of three types of entity resources: computing resources, storage resources, and transmission resources.
  • Part of the status data in the business disk of the physical space is processed on the computing storage disk, and the other part is reported to the virtual space through the device status interface S-flow.
  • the virtual space can grasp the performance status of the three types of resources in the physical space in real time, including computing resources, storage resources, and transmission resources, so that optimal decisions can be made when allocating resources.
  • the computing, storage, and transmission resource capabilities of the virtual space are uniformly described in the three-layer three-sided network virtualization model and/or the seventh-order tensor model, and are presented to the application space through the cloud network capability interface C-flow.
  • the application space is provided for various business scenarios.
  • the method proposed in this embodiment realizes the deep integration of computing, storage, and transmission resources in three types of spaces: physical, virtual, and user.
  • the most suitable integration method is selected according to business requirements. For example, in order to provide users with computing and storage services more quickly, physical space integration can be used to access communication equipment at the edge, and the computing and storage unit disks can be embedded in the communication equipment. This can not only reduce communication delays, but also reduce communication security risks.
  • virtual space resource fusion technology can be used to efficiently orchestrate services and schedule virtual resources according to user priorities.
  • the three fusion methods provided by the present invention can realize efficient and flexible fusion of computing, storage and transmission resources, and can intelligently and dynamically schedule resources to meet user needs.
  • this Embodiment 6 proposes an automatic resource integration and scheduling method based on virtualization technology.
  • the method in this embodiment includes five steps.
  • the following uses a distributed machine learning application scenario as an example to illustrate the specific implementation of the five steps.
  • the first step is to make a request and submit the request description in the application space.
  • the user makes a distributed machine learning request, submits a request description in the application space, and submits machine learning training data, which is selected as needed.
  • the application space analyzes user requests to obtain quantitative indicators of computing, storage, and transmission resources, which are described in the three-layer three-sided network virtualization model and/or the seventh-order tensor model.
  • the requirements for computing, storage, and transmission resources in a distributed machine learning application scenario are as shown in Table 4, the user training time is about 2 hours, the hard disk storage space required for training data is 2T, and the memory requirement is 256G.
  • Distributed machine learning uses data In the central cluster mode, 100Gbps bandwidth is required between the server and the switch, and the transmission delay needs to be within 20 milliseconds.
  • the five indicators in Table 4 will correspond to the dimensions of the fifth, sixth, and seventh orders of the tensor model.
  • the virtual space executes the third step, and its orchestrator and controller convert the resource quantification indicators into specific operation instructions.
  • the fourth step is to create a virtual model based on virtualization technology and instantiate it, verify the operation instructions, and issue the device after it is accurate.
  • the virtual model and the physical entity interact through the device status interface S-flow to synchronize the current status of the device in real time. Therefore, if the operation instruction cannot be executed on the physical entity, the virtual entity will fail the verification and will not issue the operation instruction to the device.
  • the computing, storage, and transmission equipment receive and execute various operation instructions.
  • the controller will find a path that meets the latency requirements, deliver the data to the corresponding server through this path, and the server will provide storage space and memory that meets the requirements, create a virtual machine, and perform machine learning. Training. If it is necessary to synchronize the intermediate parameters of the artificial neural network model during the training process, the controller will allocate a transmission path. After training, the execution results are returned to the user.
  • five types of information are exchanged between the five steps, namely user requests, resource quantification indicators and capabilities, operations instructions executed on the virtual device, and instructions executed on the computing, storage, and transmission physical devices. , execution result feedback information Feedback.
  • the five steps and five types of interactive information form a closed loop that runs through physical space, virtual space, and user space to realize automatic integration and scheduling of computing, storage, and transmission resources.
  • the automatic resource integration and scheduling method based on virtualization technology proposed in this embodiment uses an intelligent mechanism to realize virtualization, efficient integration, automatic scheduling, and feedback optimization of computing, storage, and transmission resources.
  • this intelligent automatic mechanism can solve the problems of long cycle and low efficiency caused by manual scheduling.
  • it ensures that the resource allocation results meet the business needs of the application space through the closed-loop interaction mechanism. It also ensures that the resource allocation strategy is consistent with the virtual space through the simulation verification mechanism. The physical space is highly consistent.
  • this Embodiment 7 further improves the integration method of computing, storage and transmission resources in the physical space. Detailed explanation.
  • FIG. 6 it is a schematic diagram of the integration of computing, storage and transmission resources in the physical space in this embodiment.
  • this embodiment includes three processes.
  • one or more computing storage disks are set up in the physical space.
  • the physical space also includes a main control disk and several business disks.
  • this embodiment requires one or more computing storage disks, and the specific number can be determined according to the business scenario. For example, if the transmission device contains multiple service disks and the generated device status data will be relatively large, multiple computing storage disks need to be developed and deployed. If the transmission device contains a relatively small number of business disks and the amount of device status data generated is not large, you can develop and deploy a computing storage disk. After the communication network is built and opened, the business volume will increase or decrease, but the amount of equipment status data reported by the business disk will not change. Therefore, when the communication equipment is delivered, the number of required computing storage disks can already be determined.
  • a computing storage disk with higher performance can be developed to replace it.
  • the current equipment performance indicator data collection cycle is to collect and report once every five minutes. Due to network optimization needs, it needs to be adjusted to collect and report once every one minute.
  • the currently configured computing storage disk processing capacity and storage capacity cannot meet the needs, then you can Develop higher-performance computing storage disks for replacement and upgrade.
  • the second process is implemented, as shown on the left side of Figure 6.
  • Basic software is installed and deployed, including installing the operating system (such as CentOS operating system, Ubuntu operating system), and installing data.
  • Tool software such as database MySQL, distributed file system
  • the basic software shown on the left side of Figure 6 is divided into hierarchical relationships, and lower-level software is called by upper-level software.
  • data tool software calls functions provided by the operating system to access the central processor, memory, disk, peripheral devices, etc.
  • the digital twin modeling framework, intelligent training inference framework, and algorithm analysis and processing framework call data tool software to obtain various types of data.
  • the third process is to implement data offloading and edge processing, as shown on the right side of Figure 6.
  • Business disk 1, business disk 2, ..., and business disk n collect device status data. If the collected data is closely related to this network element, it is uploaded to the computing storage disk for edge processing. If it is related to other network elements, it is uploaded to the main control disk, which then transfers it to the management and control platform. After the computing storage disk receives the device status data, it is deployed according to the data processing requirements. If the simulation modeling operation is performed, the data is forwarded to the digital twin modeling framework; if edge artificial intelligence inference is performed, the data is forwarded to intelligent training inference. Framework; if operations such as alarm correlation analysis or performance degradation analysis are performed, the data will be forwarded to the algorithm analysis and processing framework.
  • the software developed based on the above three processes is deployed on business disks, main control disks, and computing storage disks. These software adopts a distributed collaboration model to realize data offloading and task processing, and integrates with various single disks, operating systems, and data in the physical space. Tools and various frameworks work together to achieve the integration of computing, storage and transmission resources.
  • the physical space resource fusion method proposed in this embodiment can provide a new and more efficient product solution.
  • a computing storage disk is added to provide computing, storage, and storage for the upper virtual space and application space through physical space fusion. Delivery infrastructure capabilities.
  • This new product solution can provide users with cloud computing and network capabilities at the edge of the network. Cloud computing capabilities are borne by computing storage disks, and network capabilities are borne by business disks. Users' application requirements are submitted uniformly through the network.
  • the business disk analyzes and processes the data. If the user needs low-latency processing and the data scale is not particularly large, the business disk can Offload data directly to computing storage disks.
  • the business disk will divert the data to the main control disk, and then upload it to the management and control platform through the main control disk and hand it over to the virtual space for processing in a large-scale computing cluster.
  • many individual user applications have characteristics such as large number, small data, high latency requirements, and high security requirements.
  • Information infrastructure equipment is required to process data in real time on the access side. On the one hand, it can reduce latency; on the other hand, it can also reduce Security risks caused by network communications.
  • the physical space integration method proposed in this embodiment effectively solves this problem by deploying computing storage disks on communication equipment, adding software tools, and using business disks, computing storage disks, main control disks, and management and control platforms to jointly distribute data. Come up with new product solutions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

本发明涉及一种网络虚拟化体系结构以及虚拟化方法。其网络虚拟化体系结构部分主要包括物理空间、虚拟空间和应用空间，其中：物理空间内包括计算资源、存储资源、传送资源，并实现计算存储传送资源的融合，物理空间还将设备资源状态数据上报到所述虚拟空间；虚拟空间实现各项资源的虚拟化并将其描述在虚拟化模型中，以呈现给应用空间；虚拟空间还在接到应用空间下发的需求后将需求指令下发到具体的物理设备或虚拟设备上执行；应用空间获取虚拟空间描述的虚拟化模型，并提供给各类业务场景；应用空间还接纳需求、整合需求并下发给虚拟空间。本发明可以支持计算存储传送资源深度融合，支持资源全局最优统筹调度。

Description

一种网络虚拟化体系结构以及虚拟化方法 技术领域
本发明涉及通信技术领域，特别是涉及一种网络虚拟化体系结构以及虚拟化方法。
背景技术
数字经济的发展需要信息技术和通信技术支撑，行业数字化转型需求也推动传统通信网络架构向“云网融合”架构演进。传统“物理设备加专业网管”模式构建的通信网络架构很难满足各行业“资源随需、管控灵活、安全可靠”的数字化转型需求。基于虚拟化技术，构建计算、存储、传送资源池，设计新型云网融合架构，引进人工智能技术实现全网业务编排、端到端资源高效协同控制，基于内生安全技术保障虚拟资源池的安全可靠，满足各行业动态多样化业务需求，成为网络架构演进方向。
近年来，产业界和学术界针对“云网融合”技术展开了一系列研究。截止目前，相关研究成果能够支持“云网协同”，但是仍旧缺乏一种体系结构用来支撑云网融合。信息技术包括两个较大的技术体系，一是以云计算为代表的信息处理技术体系，另一个是以网络为代表的信息传输技术体系。信息处理技术专注于研究高效的算法实现大规模高性能数据处理，从数据中挖掘出有价值的信息，这类技术包括分布式计算、并行算法、数据挖掘、知识发现等。信息传输技术专注于研究连接协议和高性能器件，支持更高速率、更大容量、更远距离的数据报文传输，这类技术包括网络协议、数字信号处理、光放大技术、纠错编码技术等。信息处理技术体系的学者和工程师偏向于完善网络技术，使得网络技术能够支持并行分布式数据处理，从而为用户提供更高效的计算和存储资源。信息传输技术体系的学者和工程师偏向于增强计算存储硬件和软件，从而支撑更稳定更强大的通信网络建设。上述两种云网协同方式可以抽象概括为“完善传送资源，让网更好地协同云”、“增强计算存储资源，让云更好地协同网”。上述两种协同方式都没有将计算存储传送资源放到平等地位，也导致当前是一 种浅层次的协同，云和网仍旧处于隔离状态。这种隔离状态导致用户在使用计算存储传送资源时，仍旧需要分别提出两类资源请求，第一类是向云计算中心提出计算存储资源资求，第二类是向通信网络提出传送资源请求。这种浅层次云网协同技术有两个明显缺点。首先，用户需要同时协调两类资源的技术细节指标，导致用户难以专注于业务流程。其次，无论是“网更好地协同云”还是“云更好地协同网”都导致云网资源难以实现全局统筹调度，计算存储传送资源在局部达到最优利用率，但是仍然无法保障全局最优。
真正实现“云网融合”需要将云网资源放在同等地位，构建一种高效的体系结构来量化描述计算存储传送资源。构建这样一种体系结构面临着两大困难。首先，不能完全抛弃当前已经广泛应用在云计算中心、通信网络中的技术，这样无法实现云网平滑演进，也会造成已建资源的巨大浪费。其次，也不能割裂地对云网资源进行完善增强，然后在独立的云计算中心和通信网络之上部署全局控制器来实现计算存储传送浅层次的协同。当前相关技术研究没有从全局角度出发构建一种支撑云网资源深度融合的体系结构，也未提出高效的处理方法实现计算存储传送资源的统一量化描述。
鉴于以上情况，如何克服现有技术所存在的缺陷，解决异构资源融合难、全网协同程度弱、资源分配效率差的技术问题，是本技术领域待解决的难题。
发明内容
针对现有技术的以上缺陷或改进需求，本发明提供一种网络虚拟化体系结构以及虚拟化方法，全面分析总结云计算中心、通信网络相关技术，以网络虚拟技术为基础构建一种新型的体系结构，统一描述计算存储传送资源，并支持资源在物理空间、虚拟空间、应用空间的融合。本发明提出的网络虚拟化体系结构既符合当前云网架构平滑演进的技术趋势，也可以支持计算存储传送资源深度融合，同时支持用户在统一量化空间内同时申请所需要的资源，支持资源全局最优统筹调度。
本发明实施例采用如下技术方案：
第一方面，本发明提供了一种网络虚拟化体系结构，包括物理空间、 虚拟空间和应用空间，其中：
所述物理空间内包括计算资源、存储资源、传送资源，并实现计算存储传送资源的融合，所述物理空间还将设备资源状态数据上报到所述虚拟空间；
所述虚拟空间实现各项资源的虚拟化并将其描述在虚拟化模型中，以呈现给应用空间；所述虚拟空间还在接到应用空间下发的需求后将需求指令下发到具体的物理设备或虚拟设备上执行；
所述应用空间获取虚拟空间描述的虚拟化模型，并提供给各类业务场景；所述应用空间还接纳需求、整合需求并下发给虚拟空间。
进一步的，所述虚拟空间内包括编排器、控制器以及与物理空间相对应的虚拟计算资源、虚拟存储资源、虚拟传送资源；所述应用空间内包括用户应用、计算能力需求、存储能力需求、传送能力需求；所述物理空间与所述虚拟空间的接口包括设备状态接口、管控指令接口；所述应用空间与所述虚拟空间的接口包括云网能力接口、用户需求接口。
进一步的，所述物理空间、虚拟空间、应用空间和设备状态接口、管控指令接口、云网能力接口、用户需求接口组成闭环，具体的：
所述应用空间内实现计算能力需求、存储能力需求、传送能力需求融合，用户提出资源需求以后，由应用空间进行匹配，找到对应的虚拟资源，通过用户需求接口发送至虚拟空间；
所述虚拟空间实现虚拟计算资源、虚拟存储资源、虚拟传送资源融合，在接到用户需求以后，通过编排器和控制器进行处理，形成具体操作指令，通过管控指令接口下发到具体物理设备或者在虚拟设备上执行；
所述物理空间实现计算资源、存储资源、传送资源这三类实体资源的融合，物理空间的业务盘内的状态数据一部分在计算存储盘处理，另一部分通过设备状态接口上报到虚拟空间；
通过上报的设备资源状态数据，虚拟空间能够实时掌握物理空间内计算资源、存储资源、传送资源的性能状况，虚拟空间的计算、存储、传送资源能力统一描述在虚拟化模型中，并通过云网能力接口呈现给应用空间，由应用空间提供给各类业务场景。
进一步的，所述物理空间、所述虚拟空间、所述用户空间实现计算存 储传送资源自动融合调度的过程包括：
提出请求，在应用空间提交请求描述；
应用空间分析请求，得到计算、存储、传送资源量化指标，在虚拟化模型中描述；
完成资源量化指标统一描述以后，虚拟空间的编排器和控制器将资源量化指标转换为具体操作指令；
基于虚拟化技术创建虚拟模型并进行实例化，对操作指令进行验证，准确无误后下发设备；
计算、存储、传送设备接收并执行各项操作指令。
进一步的，所述物理空间实现计算资源、存储资源、传送资源这三类实体资源的融合过程具体包括：
在物理空间内设置一块或多块计算存储盘，所述物理空间内还包括主控盘以及若干个业务盘；
安装和部署基础软件，包括操作系统、数据工具软件、数字孪生建模框架、智能训练推理框架、算法分析处理框架中的一种或多种；其中，数据工具软件调用操作系统提供的函数访问中央处理器、内存、磁盘、外围设备中的一种或多种；数字孪生建模框架、智能训练推理框架、算法分析处理框架调用数据工具软件获取各类数据；
所述业务盘用于采集设备状态数据，若采集的数据与本网元相关，则上传到计算存储盘进行边缘处理；若与其它网元相关，则上传到主控盘，由主控盘转交至管控平台；计算存储盘收到设备状态数据后，根据数据处理需求进行调配，若进行仿真建模操作，则将数据转发给数字孪生建模框架；若进行边缘人工智能推理，则将数据转发给智能训练推理框架；若进行告警关联分析或者性能劣化分析操作，则将数据转发给算法分析处理框架。
进一步的，所述虚拟化模型包括三层三面网络虚拟化模型以及七阶张量模型，其中：
所述三层三面网络虚拟化模型包括三类资源、三层空间以及三项技术，所述三类资源包括计算资源、存储资源、传送资源，所述三层空间包括物理空间、虚拟空间、应用空间，所述三项技术包括虚拟化技术、智能运维 技术、内生安全技术；所述三类资源位于所述三层空间之内，所述三项技术作用于所述三层空间以及所述三类资源；
所述七阶张量模型对位于三层空间的三类资源进行统一描述和量化表示，所述三类技术用于对所述七阶张量模型中的各项元素进行处理，包括：基于虚拟化技术实现物理空间计算资源、存储资源、传送资源的虚拟化，基于智能运维技术实现虚拟化资源的跨域统筹调度，基于内生安全技术构建安全可靠的虚拟化体系。
进一步的，所述七阶张量模型描述为T∈R I1×I2×I3×I4×I5×I6×I7，其中R表示实数域，I1、I2、I3、I4、I5、I6、I7表示张量模型的七个阶，分别表示时间、物理空间、虚拟空间、应用空间、计算资源、存储资源、传送资源。
第二方面，本发明还提供了一种虚拟化方法，采用虚拟化技术对各项资源进行虚拟化，其中，所述虚拟化技术包括功能保真技术和功能仿真技术，所述功能保真技术包括调度保真技术、虚拟保真技术，所述功能仿真技术包括功能映射仿真技术、功能拟合仿真技术；对于一个具体的物理实体资源，在选择虚拟化技术时按照调度保真技术、虚拟保真技术、功能映射仿真技术、功能拟合仿真技术的优先顺序进行选择。
进一步的，在选择虚拟化技术时还包括P跃迁、F跃迁、FP跃迁三种跃迁方式，具体的：
若不了解某物理实体的运行机理，则需要采用功能拟合仿真技术进行虚拟化；当能够完全掌握该物理实体的运行机理时，则通过P跃迁的方式，采用功能映射仿真技术进行虚拟化；当能够让基于功能映射仿真技术构建的虚拟化实体真正承担物理实体的功能时，则通过F跃迁的方式，采用虚拟保真技术进行虚拟化；若同时满足P跃迁和F跃迁的条件，则通过FP跃迁的方式，采用调度保真技术进行虚拟化。
进一步的，在对物理实体进行虚拟化建模时，还包括对构建的虚拟化实例进行优化，具体的：
明确真实网络中部署的物理实体，将物理实体的设备状态数据、物理实体的配置数据上传到虚拟空间；
根据物理实体上传的相关设备状态数据、配置数据，梳理实体运行机 理，选择相应的虚拟化技术构建虚拟化模型；
根据应用空间业务场景需求，调用虚拟化模型，传入实时数据，创建虚拟化实例，构建虚拟网络；
应用空间执行完各类应用场景具体任务以后，反馈虚拟化效果，根据虚拟化效果对虚拟化模型进行优化。
与现有技术相比，本发明的有益效果在于：
(1)、本发明提出新型的网络虚拟化体系结构，实现计算资源、存储资源、传送资源的跨域统一调度和控制，支持业务端到端全局编排，支持资源智能运营和维护，通过一套整体的网络虚拟化体系结构实现云计算环境与通信网络的深度融合。本发明采用新型的基于张量的三层三面网络虚拟化模型以及七阶张量模型，在高阶多维度张量空间中统一描述计算、存储、传送资源，清晰地界定物理空间、虚拟空间、应用空间，引入虚拟化技术、智能运维技术、内生安全技术，解决当前网络体系架构资源随需分配能力差、资源跨域协同难、资源运维费时低效的问题。
(2)、本发明提出两大类四小类网络虚拟化技术，针对不同物理实体的物征，选择合适的虚拟化技术，可以构建出高精确度虚拟化模型。基于这些虚拟化模型，面向各应用空间用户需求，创建虚拟化网络并进行业务验证或资源运营，可以实现网络业务自动处理、网络资源自动维护、网络性能自动优化。本发明还采用虚拟化实例自动优化机制，通过物理空间、虚拟空间、应用空间跨层协同反馈技术优化虚拟化实例，提升虚拟化模型对物理实体的仿真度。
(3)、本发明提出的计算存储传送资源融合方法，能够分别在物理空间、虚拟空间、应用空间实现资源深度融合。物理空间支持数据在计算存储单盘和传送业务单盘之间的分流。虚拟空间部署业务编排器和资源控制器，支持全局业务编排和跨域资源控制，支持虚拟化实例创建、组合、优化，并通过虚拟化实例实现网络能力随需开放。在应用空间，用户无须了解太多的技术背景就可以实现计算存储传送资源的简洁精准请求，让用户更加关注业务，提升用户感知，也可以确保资源请求的准确性和一致性。
附图说明
为了更清楚地说明本发明实施例的技术方案，下面将对本发明实施例中所需要使用的附图作简单地介绍。显而易见地，下面所描述的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其它的附图。
图1为本发明实施例1提供的一种面向计算存储传送资源融合的网络虚拟化体系结构示意图；
图2为本发明实施例3提供的网络虚拟化技术树示意图；
图3为本发明实施例4提供的物理实体虚拟化建模及虚拟化实例的优化示意图；
图4为本发明实施例5提供的计算存储传送资源在物理虚拟应用空间的融合示意图；
图5为本发明实施例6提供的基于虚拟化技术的资源自动融合调度示意图；
图6为本发明实施例7提供的物理空间内计算存储传送资源的融合示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本发明，并不用于限定本发明。
本发明是一种特定功能系统的体系结构，因此在具体实施例中主要说明各结构模组的功能逻辑关系，并不对具体软件和硬件实施方式做限定。
此外，下面所描述的本发明各个实施方式中所涉及到的技术特征只要彼此之间未构成冲突就可以相互组合。下面就参考附图和实施例结合来详细说明本发明。
实施例1：
“云网融合”的核心是实现云计算中心资源与网络资源的融合。云计算资源包括计算资源和存储资源，像中央处理器、人工智能芯片等都属于计算资源。存储阵列、硬盘、内存等都属于存储资源。网络资源主要指传送资源，包括光纤、网线、通信设备单盘、光模块、光器件、网络协议等。 换句话说，“云网融合”的核心是实现计算、存储、传送三类资源的融合，而“云网融合”目标是高效、灵活、智能、安全地为用户提供计算、存储、传送资源。在当前云计算中心和通信网络中，计算、存储、传送三类资源以软件和硬件实体形态存在于物理空间中。而各行业的数字化应用是采用抽象能力的形式在用户空间使用计算、存储、传送资源的。将物理空间中的实体资源进行转换，以计算能力、存储能力、传送能力的形式提供给用户，就需要在虚拟空间基于虚拟化技术对计算、存储、传送资源进行仿真建模，并部署编排器和控制器进行调度。虚拟化技术实现实体资源的抽象建模，在用户空间将资源高效、灵活地提供给用户，但也带来了资源运维复杂性和安全性的问题，需要引入智能运维技术、内生安全技术进行支撑。
基于上述描述，以及针对传统通信网络面临的“异构资源融合难、全网协同程度弱、资源分配效率差”的问题，本发明实施例提出一种新型的网络虚拟化体系结构，实现计算资源、存储资源、传送资源的深度融合和协同调度，支持上述三类资源在物理空间、虚拟空间、应用空间的融合和调度，基于虚拟化技术、智能运维、内生安全技术实现资源高效、灵活、智能、安全供应目标。
本发明实施例1提供一种面向计算存储传送资源融合的网络虚拟化体系结构，其包括物理空间、虚拟空间和应用空间，其中：所述物理空间实现计算存储传送资源的融合，并将设备资源状态数据上报到所述虚拟空间；所述虚拟空间实现各项资源的虚拟化并将其描述在虚拟化模型中，以呈现给应用空间；所述虚拟空间还在接到应用空间下发的需求后将需求指令下发到具体的物理设备或虚拟设备上执行；所述应用空间获取虚拟空间描述的虚拟化模型，并提供给各类业务场景；所述应用空间还接纳需求、整合需求并下发给虚拟空间。
本优选实施例提出二类虚拟化技术，即功能保真虚拟化技术(功能保真技术)和功能仿真虚拟化技术(功能仿真技术)，这两者即为所述虚拟空间实现各项资源的虚拟化所采用的虚拟化技术。基于功能保真技术构建的虚拟化模型，其输入和输出与物理实体模型一致，内部处理机制与物理实体一致，只是通过软件的方法来仿真模拟硬件操作。基于功能仿真技术构建的虚拟化模型，其输入和输出与物理实体模型并不完全一致，主要是 通过软件去仿真逼近物理实体的运行方式，用于验证、监控、优化物理实体功能和性能。
本实施例提出的功能保真虚拟化技术又细分为两种具体技术，即调度保真技术和虚拟保真技术。调度保真技术是在物理网络中虚拟出多个逻辑网络切片，每个切片满足特定类型的业务需求。虚拟保真技术是将基于专用硬件的网元功能虚拟化后，部署在通用硬件上，形成虚拟网元池。
本实施例提出的功能仿真虚拟化技术也细分为两种具体的技术，即功能映射仿真技术和功能拟合仿真技术。功能映射仿真技术对物理空间网络实体的通信机理进行翻译和映射以后，在虚拟空间构建对应数字孪生模型。功能拟合仿真技术主要针对难以准确描述通信机理的网络功能，基于输入输出数据进行拟合，构建数字孪生模型。在具体实施过程中，可以根据物理实体的特征从上述虚拟化技术中进行选择，创建虚拟化资源。
如图1所示，本实施例中的虚拟化模型包括三层三面网络虚拟化模型以及七阶张量模型。具体的，基于张量的三层三面网络虚拟化模型(如图1左边所示)，英文全称为Tensor based Three Layer and Three Plane Model，英文简写为TL 3P 3M，该三层三面网络虚拟化模型包括三类资源、三层空间以及三项技术，所述三类资源位于所述三层空间之内，所述三项技术同时作用于所述三层空间以及所述三类资源，其中，所述三类资源包括计算资源、存储资源、传送资源，所述三层空间包括物理空间、虚拟空间、应用空间，所述三项技术包括虚拟化技术、智能运维技术、内生安全技术，本发明实施例基于虚拟化技术实现物理空间计算资源、存储资源、传送资源虚拟化，基于智能运维技术实现虚拟化资源的跨域统筹调度，基于内生安全技术构建稳定可靠的虚拟化体系。
本实施例的网络虚拟化体系结构还包括和三层三面网络虚拟化模型相对应的七阶张量模型(如图1右边所示)，所述七阶张量模型对位于三层空间的三类资源进行统一描述和量化表示，所述三类技术用于对所述七阶张量模型中的各项元素进行处理。包括：基于虚拟化技术实现物理空间计算资源、存储资源、传送资源的虚拟化，基于智能运维技术实现虚拟化资源的跨域统筹调度，基于内生安全技术构建安全可靠的虚拟化体系。
在本优选实施例中，七阶张量模型描述为T∈R I1×I2×I3×I4×I5×I6×I7，其 中R表示实数域，I1、I2、I3、I4、I5、I6、I7表示张量模型的七个阶，分别表示时间、物理空间、虚拟空间、应用空间、计算资源、存储资源、传送资源。
本优选实施例还提出三种资源融合方法，包括物理空间资源融合方法、虚拟空间资源融合方法、应用空间资源融合方法。传统通信设备包含许多业务单盘，只具备单一的传送功能。本优选实施例提出物理空间资源融合方法，在通信设备内增加计算存储单盘，业务单盘数据一部分上传计算存储单盘，在设备侧直接处理，一部分上传管控平台。本优选实施例提出的虚拟空间资源融合方法实现虚拟计算资源、虚拟存储资源、虚拟传送资源的跨域统筹调度，能够根据全局业务编排器下发的模板优化调度各类虚拟资源。本优选实施例提出的应用空间资源融合方法能够让行业用户在不用具备底层资源能力技术背景的情况下，直接基于业务需要请求计算存储传送资源，网络虚拟化体系结构作为一个整体将用户需求转换为资源需求，并基于虚拟化资源池进行调度分配模拟验证，准确无误以后通过控制器发下发对应的物理实体设备，实现计算存储传送资源的分配。
基于本实施例提出的上述融合方法，通信设备制造商在研发生产物理实体设备的同时，研发生产虚拟化模型，包括虚拟化计算资源模型、虚拟化存储资源模型、虚拟化传送资源模型。这些虚拟化模型由网络虚拟化体系结构统一管理控制，在运营商通信网络部署以后，物理实体实时采集上报的数据与虚拟化模型融合，形成一系列虚拟化实例，虚拟空间内的虚拟化实例与物理空间内的实体设备相对应，支持物理实体设备运行机理和健康状态实时呈现、历史回溯、前瞻预测。基于本实施例提出的方法，物理实体设备出现性能劣化等情况以后，通信网络管控平台基于智能化技术创建性能优化方案，在虚拟空间中创建虚拟化实例对性能优化方案进行验证，如果能够达到优化效果就可以直接将操作指令下发物理实体设备。另外，通信网络中很多操作花费时间比较长，但是一旦出现网络故障或性能劣化现象，需要网络能够快速恢复。基于本实施例提出的方法，虚拟化实例与现有网络健康状态精准匹配，可基于虚拟化实例预测网络潜在隐患或劣化情况，形成对应处理方案并验证，当通信网络真正出现问题的时间，可以直接采用已经准备好的处理方案，节省方案创建时间，确保通信网络实时 排障或优化。
实施例2：
基于实施例1提供的一种面向计算存储传送资源融合的网络虚拟化体系结构，本实施例2对该网络虚拟化体系结构的三层三面网络虚拟化模型以及七阶张量模型进行更为详细具体的说明。
继续如图1所示，对于本实施例的三层三面网络虚拟化模型，其整个模型呈现为一个立方体，立方体的正前方是三层，最下边一层是物理空间，部署实体设备，包括计算设备(例如服务器)、存储设备(例如磁盘阵列)、传送设备(例如光传送设备、数据通信设备)。中间一层是虚拟空间，部署业务编排器、跨域控制器、单域控制器、基于虚拟化技术创建的虚拟计算资源、虚拟存储资源、虚拟传送资源。最上层是应用空间，包含一系列用户应用，例如分布式机器学习。立方体右边包括三个面，这三个面对应三项技术，从前向后分别为虚拟化技术、智能运维技术、内生安全技术。这三项技术从上到下贯穿三层，表示三项技术同时作用于物理空间、虚拟空间、应用空间。虚拟化技术对物理空间各种实体设备运行机理进行仿真建模，构建对应的虚拟资源并部署于虚拟空间。虚拟化技术对应用空间各类业务场景建模，抽象各类应用的主要特征，根据主要特征为用户推荐高效的资源分配模式。智能运维技术在时间维度上能够实现各类资源规划阶段、开通阶段、维护阶段的自动化处理，基于人工神经网络、机器学习、知识图谱等技术实现业务编排和资源控制的智能化处理，支持网络根源与衍生告警智能关联分析、网络故障智能定位排除、性能劣化趋势智能预测。计算存储传送资源融合过程中不可避免地面临安全问题，包括各类外部入侵攻击问题，也包括内部软件硬件漏洞导致的数据安全问题，针对这些安全问题，本发明基于内生安全技术构建安全可靠的网络虚拟化体系结构，保障物理空间、虚拟空间、应用空间各类物理虚拟资源和业务场景的安全。
图1右边为本发明实施例提出的七阶张量网络虚拟化体系结构模型(七阶张量模型)，这个模型能够将三层三面网络虚拟化模型中各项要素进行量化表示，建立各项要素之间的关系。在本实施例中，位于三个空间(即物理空间、虚拟空间、用户空间)的三类资源(即计算资源、存储资源、传送资源)在七阶张量模型中统一描述和量化表示，而三类技术(即虚拟 化技术、智能运维技术、内生安全技术)则对张量模型中的元素进行处理，实现高效、灵活、智能、安全的资源供应。
具体的，本实施例的七阶张量模型描述为T∈R I1×I2×I3×I4×I5×I6×I7，其中R表示实数域，I1、I2、I3、I4、I5、I6、I7表示张量模型的七个阶，分别表示时间、物理空间、虚拟空间、应用空间、计算资源、存储资源、传送资源。张量每一阶包含很多维度，七阶七个维度就可以确定七阶张量模型中对象的具体位置，这个位置的张量元素值是一个实数，实数值对应于对象本身的具体特征值。
下面的表1以机器学习应用场景、控制器系统、通信设备计算单盘举例阐述七阶张量模型的张量元素如何取值。
表1七阶张量模型张量元素数值举例
序号 名称 I1阶 I2阶 I3阶 I4阶 I5阶 I6阶 I7阶 数值
1 机器学习 2 0 0 1 8 2 4 1
2 控制器 3 0 1 0 4 1 1 2
3 计算单盘 5 1 0 0 32 64 0 2
在上面表1中，第二行至第四行表示三个对象，第三列至第九列表示张量模型的七个阶，第十列是对象具体数值。本实施例提出的七阶张量模型是一个通用的模型，可以根据实际应用需求来定义张量七阶各维度值表征具体指标。例如，当前通信网络物理设备采用分钟级数据上报模式，每十五分钟上报一次全量性能数据，则可以指定张量模型第一阶维度值每一个整数表征一个长度为十五分钟的周期。将七阶张量模型内所有对象按起始运行时间排序，将最早运行时间对应为张量模型第一阶维度零坐标，然后维度每增加一个数字，代表跨越一个周期，即维度数值“1”表示第十五分钟，维度数值“2”表示第三十分钟。按上述规则，表1中的三个对象：机器学习、控制器、计算单盘对应时间分别为第三十分钟、第四十五分钟、第一小时十五分钟。对象机器学习是位于应用空间的业务场景，所以I4阶数值为1，I2和I3阶数值都为零。在实施过程中，可以根据具体需求定义物理空间、虚拟空间、应用空间三阶各个维度指值。表1中，将应用空间I4阶维度1定义为机器学习业务场景，维度2定义为虚拟现实应用场景，所以表1第2行第6列的数值为1。同理，控制器位于虚拟空间，张量I3 阶维度1表示控制器对象。表1的第4行第4列数值1表示张量模型I2阶对应计算单盘对象。张量模型I5阶表示计算资源CPU核数，表1第2行第7列的数值8表示机器学习需要的计算资源为8核CPU。同理，表1第2行第8列数值2表示机器学习需要的存储资源为2T硬盘空间，表1第2行第9列数值4表示机器学习需要的传送资源为2个网络切片。表1第2行第10列对应张量元素值，在本例中，数值1表示一个机器学习应用。表1第3行的对象为控制器，张量元素值2表示虚拟空间部署2套控制器。在具体实施过程中，部署2套控制器是采用主备模式防止控制器出现故障导致网络业务中断，一般情况下两套控制器配置的资源和运行状态都保持一致，所以两套控制在张量模型各阶所对应的维度也一样。控制器、计算单盘的其它阶与机器学习的定义同理，在此就不赘述了。
本实施例的上述表1通过举例的方式描述了在实施过程中如何将各种对象的需求对应至张量模型的各个维度上。表1第二行表示的机器学习对象在张量模型中元素数值可以形式化表示为T(2,0,0,1,8,2,4)＝1。表1第三行表示的控制器对象在张量模型中元素数值可以形式化表示为T(3,0,1,0,4,1,1)＝2。表1第四行表示的计算单盘对象在张量模型中元素数值可以形式化表示为T(5,1,0,0,32,64,0)＝2。上述张量元素数值表式化表示方式中，T表示张量，括号中七个数值分别表示七个张量阶各维度数值，等式右边是张量元素值。例如T(2,0,0,1,8,2,4)括号中第五个元素表示张量第五阶I5维度8，实际意义表示计算资源为8核中央处理单元CPU(Central Processing Unit)。表1中机器学习和控制器是计算存储传送资源需求方，机器学习对象需要消耗资源实现人工智能学习训练，控制器消耗资源执行通信网络路径计算等各类任务。表1中计算单盘是资源提供方，能够提供计算资源和存储资源。本实施例所提出的七阶张量模型能够对物理空间、虚拟空间、应用空间中各种对象所需要的计算、存储、传送资源进行具体量化描述，也能够对各种对象所能提供的资源能力数值进行量化描述，从而实现在一个张量模型中对计算、存储、传送资源需求和供应能力的统一量化描述。
构建本实施例提出的张量模型，最大的有益效果是为后续建立计算、存储、传送三类资源之间的复杂数量关系奠定基础。或者说，只有将计算、 存储、传送资源进行量化描述，在一个空间内实现统一表示，才可能建立各种资源之间的数量关系，而建立数量关系是实现资源最优调度的核心。基于本发明提出张量模型，虚拟化技术将计算、存储、传送资源抽像建模，以量化形式在统一空间内有序表示；人工智能技术基于张量空间元素建立计算、存储、传送资源之间的数量关系，实现云计算中心和通信网络“规划、建设、维护、优化、运营”全生命周期的智能化运维；内生安全技术基于张量空间建立计算、存储、传送资源之间的数量关系，清晰评估各种安全隐患可能为用户业务应用造成的影响，制定相关安全预案，一旦发生安全问题则启动安全预案来保障用户仍旧能够使用计算、存储、传送资源。
实施例3：
基于实施例1提供的一种面向计算存储传送资源融合的网络虚拟化体系结构，本实施例3对其采用的虚拟化技术进行更为详细具体的说明。
如图2所示，为本实施例提出的网络虚拟化技术树，这棵技术树包括一个树根，两个分支，四片树叶。树根代表网络虚拟化技术，两个分支表示功能保真技术和功能仿真技术。四片树叶分别为调度保真技术、虚拟保真技术、功能映射仿真技术、功能拟合仿真技术。
具体实施过程中，为了将物理空间的计算、存储、传送资源进行虚拟化，构建对应的虚拟实体，本实施例提出“保真优先、仿真次之”的方法，对于一个具体的物理实体资源，按照从左到右的顺序在四片树叶选择虚拟化技术，即按照①②③④(调度保真技术、虚拟保真技术、功能映射仿真技术、功能拟合仿真技术)的顺序选择。对于一个具体的物理实体资源，首先选择调度保真技术进行虚拟化，如果难以实施，则选择虚拟保真技术进行抽象建模。如果保真技术无法应用到该物理实体资源，则选择功能映射技术构建虚拟实体，如果还是无法实施，则选择功能拟合技术构建虚拟实体。在虚拟化技术领域，当前技术研究都是独立地开展，要么聚焦功能保真技术，要么专注于功能仿真技术，尚未有统一的虚拟化技术树清晰地表达各类虚拟化技术，这也导致一直缺乏有效的选择方法对物理实体进行虚拟化。本专利提出“保真优化、仿真次之”方法有效解决这一难题。
另外，图2中四片树叶代表的虚拟化技术在满足一定条件后可以沿着从右向左的顺序进行跃迁，即沿着④③②①的顺序跃迁。本实施例提出三 种跃迁方式，分别为P跃迁、F跃迁、FP跃迁。在实施过程中，如果不了解物理实体的运行机理，则需要采用功能拟合仿真技术进行虚拟化；当能够完全掌握该物理实体的运行机理时，则通过P跃迁的方式，采用功能映射仿真技术进行虚拟化；当能够让基于功能映射仿真技术构建的虚拟化实体真正承担物理实体的功能时，则通过F跃迁的方式，采用虚拟保真技术进行虚拟化；若同时满足P跃迁和F跃迁的条件，则通过FP跃迁的方式，采用调度保真技术进行虚拟化。例如，对于采购的光器件，一般都无法了解光器件内部运行机理。如果厂商后来采用自主研发本类光器件方式，能够完全掌握光器件运行机理，则通过P跃迁的方式，采用功能映射仿真技术进行虚拟化，从功能拟合仿真到功能映射仿真，实际上能够极大地提升虚拟化实体的仿真性能，因此称之P跃迁，P是性能英文单词Performance的第一字母。如果通过各种技术突破能够让基于功能映射仿真技术构建的虚拟化实体真正承担物理实体的功能，则通过F跃迁基于虚拟保真技术进行虚拟化，从功能映射仿真到虚拟保真技术，实际上扩展了虚拟实体的功能，让虚拟实体像物理实体一样工作，F是功能英文单词Function的第一个字母。如果性能提升和功能扩展都能够满足，则通过FP跃迁基于调度保真技术构建虚拟实体。如图2所示，本实施例提出的网络虚拟化技术树一共包括四类虚拟化技术、一种虚拟化技术选择方法、三种跃迁方法。前面阐述了一种虚拟化技术选择方法和三种跃迁方法，下边具体讲述四类虚拟化技术。
对于虚拟化技术，是将物理空间中的通信设备实体通过虚拟化技术构建对应的虚拟实体，并部署于虚拟空间。如果虚拟实体的功能与物理设备实体功能相对应，则需要采用功能保真技术。如果虚拟实体不必要完全映射物理设备实体的运行机理，则可以采用功能仿真技术。
本实施例的调度保真技术采用功能映射的方式，将用户对计算、存储、传送资源的请求进行收集和排序，制定资源调度优先级别，按优先级从高到低分配资源，满足用户需求。图2左下角是调度保真技术示意，最下边两个圆圈代表物理资源，最上边一个圆圈代表用户资源需求，中间方框表示调度器。调度器了解所有物理资源的可用情况，并根据用户提交的资源需求进行调度，为用户分配合适的资源。在图2左下角，调度器分配左边 圆圈代表的物理资源来完成用户的需求。
表2调度保真虚拟化技术时延路径举例
Figure PCTCN2023070889-appb-000001
表2以时延路径资源需求举例阐述调度保真虚拟技术。假如用户需要15毫秒内完成一次操作，其中计算时间为5毫秒，发送端与接收端数据传输时间需要小于5毫秒。基于本实施例构建的三层三面网络虚拟化模型以及七阶张量模型，查找发送端与接收端所有传送路径上的物理实体，将每条路径上物理实体处理数据的时延累加，得到本路径时延总计。例如，一条路径通过400公里光纤连接，传送时延为2毫秒。涉及两个光放大器，传送时延为0.2微秒。假设数通设备需要时延300微秒，再加上其他因素，例始色散补偿光纤，拥塞因素等，本路径传送时延大约为5毫秒。用户提出10毫秒资源需求时，调度器可以分配本路径以及对应的发送端-接收端给用户，满足用户对计算时延和传送时延的需求。
图2左下方第二个图描述了虚拟保真技术，底层三个圆圈表示物理设备，上边两个五角形表示虚拟设备。物理设备基于专用硬件构建，实现电信号和光信号处理，实现数据报文传送。虚拟设备基于通用硬件，利用集群提升处理性能，通过软件程序仿真物理设备数据传送功能。例如客户侧设备CPE(Customer Premise Equipment)通过虚拟化技术可以构建虚拟化客户侧设备vCPE(virtualized Customer Premise Equipment)，宽带接入服务器BRAS(Broadband Remote Access Server)通过虚拟化技术可以构建虚拟化宽带接入服务器vBRAS(virtualized Broadband Remote Access Server)。虚拟保真技术能够实现物理设备与虚拟设备运行机理的映射，确保物理设备与虚拟设备功能保持一致。通过虚拟保真技术构建的虚拟设备可以是盒式物理设备形态，在通用中央处理器上运行虚拟设备软件。也可以直接将虚拟设备软件部署在云数据中心虚拟机上。虚拟设备具体形态可以在实施过程中根据用户应用场景来确定。
图2下边第三个图描述了功能映射仿真技术。如果能够详细了解物理空间通信设备的内部机理，则可以通过功能映射仿真技术将内部机理翻译 为数学模型。表3以掺铒光纤放大器为阐述功能映射仿真技术实施步骤，包括七步。第一步，获取掺铒光纤放大器各项参数值，例始增益谱和功能谱。第二步，设置入射泵浦和信号光功率，第三步，计算泵浦光和信号光基模分布，第四步，计算入射泵浦和信号光子通量，光通信系统理论中有公式描述光子通量与功率之间的关系。第五步，计算能级分布，计算增益系数的横向分布。第六步是循环迭代，通过更新和判定等指令从入射泵浦和信号光功率，计算出放大器光纤末端的泵浦和信号光功率分布，本步骤输出仿真模型源代码。第七步，将仿真模型部署于虚拟化平台，采集网络数据，执行数据和模型融合操作，通过仿真模型进行计算，输出虚拟化仿真结果。
表3掺铒光纤放大器功能映射仿真
Figure PCTCN2023070889-appb-000002
图2下边最后一个图是功能拟合仿真虚拟化技术示意图。假如在某些情况下无法了解物理空间通信设备的内部机理，则可以在实验室内通过仪表测量通信设备输入和输出数据，基于功能拟合仿真技术构建模型，通过拟合函数逼近物理实体内部真实功能函数。图中通信设备物理实体包含五个功能模块，第一个功能模块执行完以后进行分支，上边的流程需要两个功能模块处理，下边的流程需要一个功能模块处理，两个分支处理完毕以后通过最后一个功能模块汇总处理，输出结果。通过功能拟合仿真技术，录入输入数据和数出数据，拟合出仿真函数，其中上边一个分支两个圆圈功能模块拟合为一个三角功能模块和一个六边形功能模块，下边分支拟合为一个六边形功能模块。这个图是一个示意图，用于形象地阐述功能拟合仿真技术处理流程。举例来说，如果进行光信噪比OSNR(Optical Signal Noise Ratio)虚拟化建模，可以在实验室通过仪表测量光功率、色散等数值作为输入，测量出光信噪比数值作为输出，选择深度神经网络DNN(Deep Neural Network)模型拟合，得到光信噪比虚拟化仿真模型。
本实施例提出的网络虚拟化技术树一共包括四类虚拟化技术、一种虚 拟化技术选择方法、三种跃迁方法，基于上述技术和方法能够最有效地实现物理实体的虚拟化，并根据情境的变换更新虚拟化实体，实现性能和功能的最优，并为后续计算存储传送资源的融合奠定基础。
实施例4：
基于实施例1提供的一种面向计算存储传送资源融合的网络虚拟化体系结构以及实施例3所描述的网络虚拟化技术，如图3所示，本实施例4提出一种物理实体虚拟化建模及虚拟化实例的优化方法，其在对物理实体进行虚拟化建模时，还包括对构建的虚拟化实例进行优化。
本实施例的方法包括四个步骤。
第一步，明确真实网络中部署的物理实体，将物理实体的设备状态数据、物理实体的配置数据等上传虚拟化空间(对应图3中的“采集”)。在图3的下部是物理实体示例，包括光传送设备和光纤，光传送设备各类单盘型号和参数，各类模块和器件型号参数都上传至虚拟空间，光纤类型、光纤色散系数、光纤衰减系数等参数都上传至虚拟空间。
第二步，根据物理实体上传的相关设备状态数据、配置数据，梳理实体运行机理，选择相应的虚拟化技术构建虚拟化模型(基于实施例3提出的网络虚拟化技术构建虚拟化模型，对应图3中的“建模”)。在图3中，如果能够准确了解物理实体的内部处理流程，可以采用功能映射仿真技术构建虚拟化模型。如果无法了解物理实体的内部处理流程，可以采用功能拟合仿真技术构建虚拟化模型。
第三步，根据用户空间业务场景需求，调用虚拟化模型，传入实时数据，创建虚拟化实例，构建虚拟网络(对应图3中的“调用”)。图3上边给出一个虚拟化网络示例，包含六个网络节点，这个虚拟化网络可以支撑分布式机器学习等应用场景。
第四步，应用空间执行完各类应用场景具体任务以后，反馈虚拟化效果，根据虚拟化效果对虚拟化模型进行优化(对应图3中的“虚拟化效果反馈”、“调整组合策略”、“优化虚拟算法”、“更新采集模式”)。虚拟化模型的优化分为三类情形，第一类情形更新数据采集模式，提高数据采集上报频率或者增加新的数据指标，提升模型虚拟化效果。第二类情形优化虚拟算法，例如增强或者完善功能映射方法，采用更优质的拟合函 数，从而达到提升仿真准确度的目标。第三类情形调整组合策略，这类情况用于复杂业务场景仿真，通过调用不同的虚拟化模型来提升模型虚拟化效果，例如增加虚拟计算资源同时减少虚拟存储资源。
通过上述四个步骤可以实现虚拟化模型和虚拟化实例的升级优化。本实施例提出的物理实体虚拟化建模及虚拟化实例升级优化方法是一种动态自动的方法。传统的方法是静态手动的，构建完虚拟化模型以后就被动等待调用，然后被动等待一下轮调整，并且调整策略比较单一。本实施例中，虚拟化模型构建完成以后，在虚拟空间生成虚拟化实例，所有实例在虚拟化空间融合以后为用户提供计算、存储、传送能力。这些能力提供给用户以后，将主动地根据用户反馈进行升级优化，优化方法包括本实施例提出的更新数据采集模式、优化虚拟算法、调整组合策略。其中优化虚拟算法在具施执行过程中可以采用本发明实施例3中的虚拟化技术跃迁方法。本实施例中的动态自动升级优化方法能够实时完善虚拟化模型，实时创建高精准度的虚拟化实例，确保虚拟空间内创建的虚拟计算、虚拟存储、虚拟传送资源能够实时满足用户的多样化业务应用需求，保障云网融合服务质量。
实施例5：
基于实施例1提供的一种面向计算存储传送资源融合的网络虚拟化体系结构，本实施例5提出计算存储传送资源的三种融合方式，分别在物理空间、虚拟空间、应用空间实现资源融合。
如图4所示为本实施例中计算存储传送资源在物理虚拟应用空间的融合示意图。
图4最下部是物理空间P(Physical Space)，部署计算、存储、传送资源的实体设备，也即物理空间内包括计算资源、存储资源、传送资源。本实施例提出的物理空间计算存储传送资源融合方法是指在传送通信设备上增加计算和存储单元盘(可简称为计算存储盘)，业务盘上的状态数据一部分上传计算存储盘，一部分上传管控平台。如果业务盘上的状态数据只与本通信设备有关，则上传计算存储盘，在设备侧进行处理。如果业务盘上的状态数据与其他网元有关联，则上传管控平台进行集中处理。计算存储盘部署人工智能学习训练框架，支持轻量级人工智能模型训练和推理。
图4中间为虚拟空间V(Virtua Space)，部署编排器和控制器，另外，还基于实施例3提出的网络虚拟化技术构建虚拟计算资源、虚拟存储资源、虚拟传送资源，这三类虚拟资源也部署于虚拟空间。虚拟资源与物理资源是对应的，实施例3提出的四类网络虚拟化技术中，基于调度保真技术、功能映射仿真技术、功能拟合仿真技术构建的虚拟资源最终执行实际指令都是在对应的物理资源上。基于虚拟保真技术构建的虚拟资源可以执行实际指令。图4中虚拟空间与物理空间有两类接口，其中设备状态通过设备状态接口S-flow(State flow)从物理空间上传至虚拟空间。管控指令通过管控指令接口M-flow(Management and control flow)从虚拟空间下发至物理空间。
图4上部为应用空间A(Application space),包含各类用户应用，例如分布式机器学习应用、云化虚拟现实应用。另外，应用空间存储用户对资源的能力需求，包括计算能力需求、存储能力需求、传送能力需求。在实施例1、2中，计算存储传送资源能力统一描述在三层三面网络虚拟化模型以及七阶张量模型中，用户计算存储传送能力需求提交以后，应用空间会搜索张量模型，匹配出所需资源并分配给用户。应用空间与虚拟空间的接口包括云网能力接口C-flow(Capability flow)，用户需求接口R-flow(Requirement flow)。
在图4中，三个空间(物理空间、虚拟空间、应用空间)和四类接口(设备状态接口、管控指令接口、云网能力接口、用户需求接口)组成一个闭环。具体的，应用空间内实现计算能力需求、存储能力需求、传送能力需求融合，用户提出资源需求以后，由应用空间进行匹配，找到对应的虚拟资源，通过用户需求接口R-flow发送至虚拟空间。虚拟空间实现虚拟计算资源、虚拟存储资源、虚拟传送资源融合，在接到用户需求以后，通过编排器和控制器进行处理，形成具体操作指令，通过管控接口M-flow下发到具体物理设备或者在虚拟设备上执行。物理空间实现计算资源、存储资源、传送资源三类实体资源的融合，物理空间的业务盘内的状态数据一部分在计算存储盘处理，另一部分通过设备状态接口S-flow上报到虚拟空间。通过上报的设备状态数据，虚拟空间能够实时掌握物理空间内计算资源、存储资源、传送资源这三类资源的性能状况，从而在调配资源时能够 实现优化决策。本实施例中，虚拟空间的计算、存储、传送资源能力统一描述在三层三面网络虚拟化模型和/或七阶张量模型中，并通过云网能力接口C-flow呈现给应用空间，由应用空间提供给各类业务场景。
本实施例提出的方法实现了计算、存储、传送资源在物理、虚拟、用户三类空间内的深度融合，在具体实施过程中根据业务需求选择最适合的融合方式。例如，为了更快速为用户提供计算存储服务，可以在边缘接入通信设备采用物理空间融合方式，将计算和存储单元盘嵌入通信设备，既能降低通信时延，也可以降低通信安全隐患，同时提供计算存储传送能力。再举另外一个例子，针对突发大规模海量计算存储传送资源需求，如果当前的资源难以满足用户需求，则可以利用虚拟空间资源融合技术，根据用户的优先级高效编排业务并调度虚拟资源。如果调度过程中出现用户需求变更，则基于人工智能技术进行动态虚拟资源调度，并通过管控接口M-flow下发物理设备。综上所述，本发明提供的三种融合方式能实现高效灵活的计算存储传送资源融合，能够智能动态调度资源以满足用户需要。
实施例6：
基于实施例1提供的一种面向计算存储传送资源融合的网络虚拟化体系结构，以及实施例5提供的三种融合方式，本实施例6提出一种基于虚拟化技术的资源自动融合调度方法。
如图5所示，本实施例的方法包括五个步骤，下面以分布式机器学习应用场景为例来阐述五个步骤的具体实施方式。
第一步，提出请求，在应用空间提交请求描述。该步中，用户提出分布式机器学习请求，在应用空间提交请求描述，提交机器学习训练数据，该数据根据需要进行选择。
第二步，应用空间分析用户请求，得到计算、存储、传送资源量化指标，在三层三面网络虚拟化模型和/或七阶张量模型中描述。假如分布式机器学习应用场景对计算、存储、传送资源的需求如表4所示，用户训练时间大约为2小时，训练数据所需硬盘存储空间为2T，内存需求256G，分布式机器学习采用数据中心集群方式，服务器与交换机之间需要100Gbps带宽，传送时延需要在20毫秒以内。表4中的五个指标将对应到张量模型第五阶、第六阶、第七阶的各个维度上。
表4分布式机器学习应用场景资源需求量化指标举例
应用名称 训练时间 硬盘存储空间 内存需求 传送带宽需求 时延需求
分布式机器学习 2小时 2T 256G 100Gbps 20毫秒
完成资源量化指标统一描述以后，虚拟空间执行第三步，其编排器和控制器将资源量化指标转换为具体操作指令。
第四步，基于虚拟化技术创建虚拟模型并进行实例化，对操作指令进行验证，准确无误后下发设备。在本步骤中，虚拟模型与物理实体通过设备状态接口S-flow交互，实时同步设备当前状态。因此，操作指令如果无法在物理实体执行，虚拟实体将验证不通过，也不会将操作指令下发设备。
第五步，计算、存储、传送设备接收并执行各项操作指令。针对表4中的资源需求，控制器将求取符合时延需求的路径，将数据通过这条路径下发至对应的服务器，服务器提供符合要求的存储空间和内存，创建虚拟机并执行机器学习训练，训练过程中如果需要同步人工神经网络模型中间参数，则由控制器分配传送路径。训练结束后将执行结果返回给用户。
在上述实施过程中，五个步骤之间交互五类信息，分别是用户请求Request，资源量化指标能力需求Capabilities，在虚拟设备上执行的操作指令Operations，在计算存储传送物理设备上执行的指令Instructions，执行结果反馈信息Feedback。五个步骤与五类交互信息构成一个闭环，贯穿物理空间、虚拟空间、用户空间，实现计算存储传送资源自动融合调度。
本实施例提出的基于虚拟化技术的资源自动融合调度方法采用一种智能机制实现计算存储传送资源的虚拟化、高效融合、自动调度、反馈优化。这个智能自动机制一方面能够解决手工调度导致的周期长、效率低等问题，另一方面通过闭环交互机制保障资源分配结果满足应用空间的业务需求，通过模拟验证机制保障资源分配策略在虚拟空间与物理空间高度一致。
实施例7：
基于实施例1提供的一种面向计算存储传送资源融合的网络虚拟化体系结构，以及实施例5提供的三种融合方式，本实施例7对物理空间内计算存储传送资源的融合方式进行更为详细的说明。
如图6所示，为本实施例中物理空间内计算存储传送资源的融合示意图。在具体实施时，本实施例包括三个过程。
首先，在物理空间内设置一块或多块计算存储盘，另外，物理空间内还包括主控盘以及若干个业务盘。具体的，本实施例需要一块或者多块计算存储盘，具体数量可以根据业务场景来确定。例如，如果传送设备包含多个业务盘，产生的设备状态数据会比较多，则需要研发并部署多块计算存储盘。如果传送设备包含的业务盘数量比较少，产生的设备状态数据量也不大，则可以研发并部署一块计算存储盘。通信网络建设开通以后，业务量会增大或者减少，但是业务盘上报的设备状态数据量一盘不会变化，所以通信设备交付的时候，所需的计算存储盘数量已经可以确定。如果后续因为设备状态数据采集频率加快导致设备状态数据量增加，则可以研发性能更高的计算存储盘来替换。例如，当前设备性能指标数据采集周期为五分钟采集并上报一次，因为网络优化需要，后续需要调整为一分钟采集并上报一次，当前配置的计算存储盘处理能力和存储能力难以满足需要，则可以研发更高性能的计算存储盘进行替换升级。
在通信设备研发并部署计算存储盘以后，则进行第二个过程的实施，如图6左边所示，安装和部署基础软件，包括安装操作系统(例如CentOS操作系统、Ubuntu操作系统)，安装数据工具软件(例如数据库MySQL、分布式文件系统)，安装数字孪生建模框架、智能训练推理框架、算法分析处理框架等。图6左边所示的基础软件分为层次关系，下层软件被上层软件调用。如：数据工具软件调用操作系统提供的函数访问中央处理器、内存、磁盘、外围设备等。数字孪生建模框架、智能训练推理框架、算法分析处理框架调用数据工具软件获取各类数据。
第三个过程是实现数据分流和边缘处理，如图6右边所示。业务盘1、业务盘2、…、业务盘n采集设备状态数据，如果采集的数据与本网元密切相关，则上传到计算存储盘进行边缘处理。如果与其它网元相关，则上传到主控盘，由主控盘转交至管控平台。计算存储盘收到设备状态数据后，根据数据处理需求进行调配，如果进行仿真建模操作，则将数据转发给数字孪生建模框架；如果进行边缘人工智能推理，则将数据转发给智能训练推理框架；如果进行告警关联分析或者性能劣化分析等操作，则将数据转发给算法分析处理框架。
基于上述三个过程研发的软件部署于业务盘、主控盘、计算存储盘， 这些软件采用分布式协同的模式实现数据分流和任务处理，并与物理空间的各类单盘、操作系统、数据工具、各类框架一起实现计算存储传送资源融合。
本实施例提出的物理空间资源融合方式可提供新的更高效的产品方案，在通信设备业务盘之外增加计算存储盘，通过物理空间融合的方式为上层虚拟空间和应用空间提供计算、存储、传送基础设施能力。这种新型产品方案可以在网络边缘接入位置为用户提供云计算和网络能力，其中云计算能力由计算存储盘承担，网络能力由业务盘承担。用户的应用需求统一通过网络提交，在新型产品方案中，基于本实施例提出的方法，业务盘对数据进行分析处理，如果用户需要低时延处理，并且数据规模不是特别大，可以由业务盘将数据直接分流至计算存储盘。如果数据规划比较大，超出计算存储盘的能力，则由业务盘将数据分流至主控盘，然后通过主控盘上传至管控平台，交给虚拟空间，在大规模计算集群中进行处理。目前很多个人用户应用表现出数量多、数据少、时延要求高、安全性要求高等特征，需要信息基础设备能够在接入侧实时处理数据，一方面可以降低时延，另一方面也能够减少网络通信带来的安全隐患。本实施例提出的物理空间融合方式通过在通信设备部署计算存储盘，增加软件工具，采用业务盘、计算存储盘、主控盘、管控平台共同分流数据的形式有效解决这一问题，并可催生出新的产品方案。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。本说明书中未作详细描述的内容属于本领域专业技术人员公知的现有技术。

Claims (10)

  1. 一种网络虚拟化体系结构，其特征在于，包括物理空间、虚拟空间和应用空间，其中：
    所述物理空间内包括计算资源、存储资源、传送资源，并实现计算存储传送资源的融合，所述物理空间还将设备资源状态数据上报到所述虚拟空间；
    所述虚拟空间实现各项资源的虚拟化并将其描述在虚拟化模型中，以呈现给应用空间；所述虚拟空间还在接到应用空间下发的需求后将需求指令下发到具体的物理设备或虚拟设备上执行；
    所述应用空间获取虚拟空间描述的虚拟化模型，并提供给各类业务场景；所述应用空间还接纳需求、整合需求并下发给虚拟空间。
  2. 根据权利要求1所述的网络虚拟化体系结构，其特征在于，所述虚拟空间内包括编排器、控制器以及与物理空间相对应的虚拟计算资源、虚拟存储资源、虚拟传送资源；所述应用空间内包括用户应用、计算能力需求、存储能力需求、传送能力需求；所述物理空间与所述虚拟空间的接口包括设备状态接口、管控指令接口；所述应用空间与所述虚拟空间的接口包括云网能力接口、用户需求接口。
  3. 根据权利要求2所述的网络虚拟化体系结构，其特征在于，所述物理空间、虚拟空间、应用空间和设备状态接口、管控指令接口、云网能力接口、用户需求接口组成闭环，具体的：
    所述应用空间内实现计算能力需求、存储能力需求、传送能力需求融合，用户提出资源需求以后，由应用空间进行匹配，找到对应的虚拟资源，通过用户需求接口发送至虚拟空间；
    所述虚拟空间实现虚拟计算资源、虚拟存储资源、虚拟传送资源融合，在接到用户需求以后，通过编排器和控制器进行处理，形成具体操作指令，通过管控指令接口下发到具体物理设备或者在虚拟设备上执行；
    所述物理空间实现计算资源、存储资源、传送资源这三类实体资源的融合，物理空间的业务盘内的状态数据一部分在计算存储盘处理，另一部分通过设备 状态接口上报到虚拟空间；
    通过上报的设备资源状态数据，虚拟空间能够实时掌握物理空间内计算资源、存储资源、传送资源的性能状况，虚拟空间的计算、存储、传送资源能力统一描述在虚拟化模型中，并通过云网能力接口呈现给应用空间，由应用空间提供给各类业务场景。
  4. 根据权利要求3所述的网络虚拟化体系结构，其特征在于，所述物理空间、所述虚拟空间、所述用户空间实现计算存储传送资源自动融合调度的过程包括：
    提出请求，在应用空间提交请求描述；
    应用空间分析请求，得到计算、存储、传送资源量化指标，在虚拟化模型中描述；
    完成资源量化指标统一描述以后，虚拟空间的编排器和控制器将资源量化指标转换为具体操作指令；
    基于虚拟化技术创建虚拟模型并进行实例化，对操作指令进行验证，准确无误后下发设备；
    计算、存储、传送设备接收并执行各项操作指令。
  5. 根据权利要求3所述的网络虚拟化体系结构，其特征在于，所述物理空间实现计算资源、存储资源、传送资源这三类实体资源的融合过程具体包括：
    在物理空间内设置一块或多块计算存储盘，所述物理空间内还包括主控盘以及若干个业务盘；
    安装和部署基础软件，包括操作系统、数据工具软件、数字孪生建模框架、智能训练推理框架、算法分析处理框架中的一种或多种；其中，数据工具软件调用操作系统提供的函数访问中央处理器、内存、磁盘、外围设备中的一种或多种；数字孪生建模框架、智能训练推理框架、算法分析处理框架调用数据工具软件获取各类数据；
    所述业务盘用于采集设备状态数据，若采集的数据与本网元相关，则上传 到计算存储盘进行边缘处理；若与其它网元相关，则上传到主控盘，由主控盘转交至管控平台；计算存储盘收到设备状态数据后，根据数据处理需求进行调配，若进行仿真建模操作，则将数据转发给数字孪生建模框架；若进行边缘人工智能推理，则将数据转发给智能训练推理框架；若进行告警关联分析或者性能劣化分析操作，则将数据转发给算法分析处理框架。
  6. 根据权利要求1-5任一所述的网络虚拟化体系结构，其特征在于，所述虚拟化模型包括三层三面网络虚拟化模型以及七阶张量模型，其中：
    所述三层三面网络虚拟化模型包括三类资源、三层空间以及三项技术，所述三类资源包括计算资源、存储资源、传送资源，所述三层空间包括物理空间、虚拟空间、应用空间，所述三项技术包括虚拟化技术、智能运维技术、内生安全技术；所述三类资源位于所述三层空间之内，所述三项技术作用于所述三层空间以及所述三类资源；
    所述七阶张量模型对位于三层空间的三类资源进行统一描述和量化表示，所述三类技术用于对所述七阶张量模型中的各项元素进行处理，包括：基于虚拟化技术实现物理空间计算资源、存储资源、传送资源的虚拟化，基于智能运维技术实现虚拟化资源的跨域统筹调度，基于内生安全技术构建安全可靠的虚拟化体系。
  7. 根据权利要求6所述的网络虚拟化体系结构，其特征在于，所述七阶张量模型描述为T∈R I1×I2×I3×I4×I5×I6×I7，其中R表示实数域，I1、I2、I3、I4、I5、I6、I7表示张量模型的七个阶，分别表示时间、物理空间、虚拟空间、应用空间、计算资源、存储资源、传送资源。
  8. 一种虚拟化方法，其特征在于，采用虚拟化技术对各项资源进行虚拟化，其中，所述虚拟化技术包括功能保真技术和功能仿真技术，所述功能保真技术包括调度保真技术、虚拟保真技术，所述功能仿真技术包括功能映射仿真技术、功能拟合仿真技术；对于一个具体的物理实体资源，在选择虚拟化技术时按照调度保真技术、虚拟保真技术、功能映射仿真技术、功能拟合仿真技术 的优先顺序进行选择。
  9. 根据权利要求8所述的虚拟化方法，其特征在于，在选择虚拟化技术时还包括P跃迁、F跃迁、FP跃迁三种跃迁方式，具体的：
    若不了解某物理实体的运行机理，则需要采用功能拟合仿真技术进行虚拟化；当能够完全掌握该物理实体的运行机理时，则通过P跃迁的方式，采用功能映射仿真技术进行虚拟化；当能够让基于功能映射仿真技术构建的虚拟化实体真正承担物理实体的功能时，则通过F跃迁的方式，采用虚拟保真技术进行虚拟化；若同时满足P跃迁和F跃迁的条件，则通过FP跃迁的方式，采用调度保真技术进行虚拟化。
  10. 根据权利要求8-9任一所述的虚拟化方法，其特征在于，在对物理实体进行虚拟化建模时，还包括对构建的虚拟化实例进行优化，具体的：
    明确真实网络中部署的物理实体，将物理实体的设备状态数据、物理实体的配置数据上传到虚拟空间；
    根据物理实体上传的相关设备状态数据、配置数据，梳理实体运行机理，选择相应的虚拟化技术构建虚拟化模型；
    根据应用空间业务场景需求，调用虚拟化模型，传入实时数据，创建虚拟化实例，构建虚拟网络；
    应用空间执行完各类应用场景具体任务以后，反馈虚拟化效果，根据虚拟化效果对虚拟化模型进行优化。
PCT/CN2023/070889 2022-03-25 2023-01-06 一种网络虚拟化体系结构以及虚拟化方法 WO2023179180A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210298990.6A CN115037590B (zh) 2022-03-25 2022-03-25 一种网络虚拟化体系结构以及虚拟化方法
CN202210298990.6 2022-03-25

Publications (1)

Publication Number Publication Date
WO2023179180A1 true WO2023179180A1 (zh) 2023-09-28

Family

ID=83119321

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/070889 WO2023179180A1 (zh) 2022-03-25 2023-01-06 一种网络虚拟化体系结构以及虚拟化方法

Country Status (2)

Country Link
CN (1) CN115037590B (zh)
WO (1) WO2023179180A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115037590B (zh) * 2022-03-25 2023-08-11 烽火通信科技股份有限公司 一种网络虚拟化体系结构以及虚拟化方法
CN115794422B (zh) * 2023-02-08 2023-06-13 中国电子科技集团公司第十研究所 一种测控基带处理池资源管控编排系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103455880A (zh) * 2013-08-29 2013-12-18 国家电网公司 一种基于虚拟化技术的电网调度自动化系统
CN103544555A (zh) * 2013-08-22 2014-01-29 国家电网公司 一种电网调度自动化系统中的统一资源管理平台
CN107193627A (zh) * 2017-03-30 2017-09-22 中国电力科学研究院 一种基于虚拟化技术的仿真场景创建方法和装置
US20180285134A1 (en) * 2017-03-31 2018-10-04 The Boeing Company Emulation of hardware components
CN109189553A (zh) * 2018-08-17 2019-01-11 烽火通信科技股份有限公司 网络服务与虚拟资源多目标匹配方法及系统
CN115037590A (zh) * 2022-03-25 2022-09-09 烽火通信科技股份有限公司 一种网络虚拟化体系结构以及虚拟化方法

Family Cites Families (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7752417B2 (en) * 2006-06-05 2010-07-06 Oracle America, Inc. Dynamic selection of memory virtualization techniques
US9112769B1 (en) * 2010-12-27 2015-08-18 Amazon Technologies, Inc. Programatically provisioning virtual networks
CN102103518B (zh) * 2011-02-23 2013-11-13 运软网络科技(上海)有限公司 一种在虚拟化环境中管理资源的系统及其实现方法
CN102739771A (zh) * 2012-04-18 2012-10-17 上海和辰信息技术有限公司 一种支持服务融合的云应用集成管理平台和方法
US9503310B1 (en) * 2012-11-27 2016-11-22 Leidos, Inc. Methods and systems of dynamic management of resources in a virtualized environment
US9979602B1 (en) * 2014-08-25 2018-05-22 Cisco Technology, Inc. Network function virtualization infrastructure pod in a network environment
US9712386B1 (en) * 2015-02-18 2017-07-18 Amazon Technologies, Inc. Grouping routing resources for isolated virtual network traffic management
US20170063952A1 (en) * 2015-08-21 2017-03-02 International Business Machines Corporation Moving a portion of a streaming application to a public cloud based on sensitive data
US11055451B2 (en) * 2015-10-28 2021-07-06 Qomplx, Inc. System and methods for multi-language abstract model creation for digital environment simulations
EP3596668A1 (en) * 2017-03-14 2020-01-22 British Telecommunications Public Limited Company Virtualised software application performance
WO2019016576A1 (en) * 2017-07-17 2019-01-24 Telefonaktiebolaget Lm Ericsson (Publ) MODULAR VIRTUALIZED FUNCTION DESIGN FOR EXTENSIBILITY AND COMPOSITION
CN107566184A (zh) * 2017-09-22 2018-01-09 天翼电子商务有限公司 一种资源统一管理方法及其系统
RU2665246C1 (ru) * 2017-11-09 2018-08-28 Российская Федерация, от имени которой выступает Государственная корпорация по космической деятельности "РОСКОСМОС" Аппаратно-вычислительный комплекс виртуализации и управления ресурсами в среде облачных вычислений
US10979318B2 (en) * 2018-02-06 2021-04-13 Oracle International Corporation Enhancing resource allocation for application deployment
CN108762768B (zh) * 2018-05-17 2021-05-18 烽火通信科技股份有限公司 网络服务智能化部署方法及系统
CN109274529A (zh) * 2018-09-04 2019-01-25 有份儿智慧科技股份有限公司 基于全息大数据使实体资源数字虚拟化的融合成型方法
CN109284280B (zh) * 2018-09-06 2020-03-24 百度在线网络技术（北京）有限公司 仿真数据优化方法、装置及存储介质
CN109714219B (zh) * 2019-03-13 2021-11-09 大连大学 一种基于卫星网络的虚拟网络功能快速映射方法
CN110011835B (zh) * 2019-03-14 2021-10-01 烽火通信科技股份有限公司 网络仿真方法及系统
CN110045608B (zh) * 2019-04-02 2022-04-05 太原理工大学 基于数字孪生的机械设备零部件结构参数动态优化方法
US11343161B2 (en) * 2019-11-04 2022-05-24 Vmware, Inc. Intelligent distributed multi-site application placement across hybrid infrastructure
CN111695734A (zh) * 2020-06-12 2020-09-22 中国科学院重庆绿色智能技术研究院 一种基于数字孪生及深度学习的多工艺规划综合评估系统及方法
CN112000421B (zh) * 2020-07-15 2023-11-17 北京计算机技术及应用研究所 基于超融合架构的管理调度技术
CN112101899B (zh) * 2020-09-09 2023-02-03 北京航空航天大学 一种数字孪生增强的制造服务信息物理融合方法
CN112487668B (zh) * 2020-12-21 2021-07-13 广东工业大学 一种基于数字孪生的近物理仿真集成调试方法及其系统
CN112800106B (zh) * 2021-01-15 2022-04-01 上海大学 一种基于数字孪生技术采用历史数据驱动虚拟模型仿真的方法
CN113543170B (zh) * 2021-03-03 2024-03-01 中国电子科技集团公司电子科学研究院 基于空间计算的卫星通信系统架构及业务应用处理方法
CN113079050B (zh) * 2021-03-31 2022-08-02 广东电网有限责任公司电力调度控制中心 网络切片下基于主动探测的虚拟网资源分配方法及装置
CN112989627B (zh) * 2021-04-16 2021-08-13 成都赢瑞科技有限公司 基于虚拟时间的多学科联合仿真系统和方法
CN114021400A (zh) * 2021-09-23 2022-02-08 中铁第一勘察设计院集团有限公司 基于数字孪生的受电弓监控运维系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103544555A (zh) * 2013-08-22 2014-01-29 国家电网公司 一种电网调度自动化系统中的统一资源管理平台
CN103455880A (zh) * 2013-08-29 2013-12-18 国家电网公司 一种基于虚拟化技术的电网调度自动化系统
CN107193627A (zh) * 2017-03-30 2017-09-22 中国电力科学研究院 一种基于虚拟化技术的仿真场景创建方法和装置
US20180285134A1 (en) * 2017-03-31 2018-10-04 The Boeing Company Emulation of hardware components
CN109189553A (zh) * 2018-08-17 2019-01-11 烽火通信科技股份有限公司 网络服务与虚拟资源多目标匹配方法及系统
CN115037590A (zh) * 2022-03-25 2022-09-09 烽火通信科技股份有限公司 一种网络虚拟化体系结构以及虚拟化方法

Also Published As

Publication number Publication date
CN115037590B (zh) 2023-08-11
CN115037590A (zh) 2022-09-09

Similar Documents

Publication Publication Date Title
WO2023179180A1 (zh) 一种网络虚拟化体系结构以及虚拟化方法
WO2023246568A1 (zh) 一种数字孪生架构、模型评价体系以及模型运行方法
CN108762768B (zh) 网络服务智能化部署方法及系统
CN111064633B (zh) 一种云边协同电力信息通信设备自动化测试资源分配方法
CN113794494B (zh) 一种面向低轨卫星网络的边缘计算系统及计算卸载优化方法
CN113326002A (zh) 基于计算迁移的云边协同控制系统及迁移决策生成方法
Mechalikh et al. PureEdgeSim: A simulation framework for performance evaluation of cloud, edge and mist computing environments
CN112600717A (zh) 基于sdn的卫星网络管控协议半实物试验装置
CN107193627A (zh) 一种基于虚拟化技术的仿真场景创建方法和装置
CN105069702B (zh) 一种电网集成信息处理方法
Patni et al. Load balancing strategies for grid computing
Rayati et al. An optimal and decentralized transactive energy system for electrical grids with high penetration of renewable energy sources
Xue et al. EdgeLD: Locally distributed deep learning inference on edge device clusters
Won et al. Astra-sim2. 0: Modeling hierarchical networks and disaggregated systems for large-model training at scale
CN115395647A (zh) 基于一致性算法的配电台区群云边协同调控方法和系统
Li et al. QoS-aware and multi-objective virtual machine dynamic scheduling for big data centers in clouds
CN108540407A (zh) 一种大数据平台中Spark Streaming接收器动态配置方法及装置
Chen et al. Data-driven task offloading method for resource-constrained terminals via unified resource model
CN105471986B (zh) 一种数据中心建设规模评估方法及装置
Zheng et al. Towards an adaptive human-centric computing resource management framework based on resource prediction and multi-objective genetic algorithm
CN110662272A (zh) 一种基于蜂群无人机的最少数量领航者选取方法
CN115543582A (zh) 超算算力网络统一调度方法、系统及设备
CN111970145B (zh) 基于语义化虚拟和任务迁移的物联网设备管控方法
CN110674935B (zh) 向机载嵌入式平台移植智能算法的方法及智能计算平台
CN113220400A (zh) 基于软件定义的云桌面系统质量控制方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23773432

Country of ref document: EP

Kind code of ref document: A1