CN111984364A - Artificial intelligence cloud platform for 5G era - Google Patents
Artificial intelligence cloud platform for 5G era Download PDFInfo
- Publication number
- CN111984364A CN111984364A CN201910426197.8A CN201910426197A CN111984364A CN 111984364 A CN111984364 A CN 111984364A CN 201910426197 A CN201910426197 A CN 201910426197A CN 111984364 A CN111984364 A CN 111984364A
- Authority
- CN
- China
- Prior art keywords
- distributed
- cloud computing
- computing
- mobile
- server
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
-
- G—PHYSICS
- G08—SIGNALLING
- G08C—TRANSMISSION SYSTEMS FOR MEASURED VALUES, CONTROL OR SIMILAR SIGNALS
- G08C17/00—Arrangements for transmitting signals characterised by the use of a wireless electrical link
- G08C17/02—Arrangements for transmitting signals characterised by the use of a wireless electrical link using a radio link
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/51—Discovery or management thereof, e.g. service location protocol [SLP] or web services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/4557—Distribution of virtual machine instances; Migration and load balancing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/45595—Network integration; Enabling network access in virtual machine instances
-
- 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
Abstract
The invention discloses an artificial intelligence cloud platform facing to the 5G era, and particularly relates to the fields of artificial intelligence, cloud computing and big data, wherein the artificial intelligence cloud platform comprises a heterogeneous distributed cloud platform, wherein the heterogeneous distributed cloud platform takes a distributed mobile cloud computing cooperative architecture as a basic architecture; the distributed mobile cloud computing cooperative architecture comprises an intelligent terminal, a mobile network, a distributed mobile cloud computing server, a cooperative controller and an end-to-end-terminal computing unloading service quality assurance mechanism system. The invention establishes a distributed mobile cloud computing cooperative framework facing 5G and an end-to-end-terminal computing unloading service quality guarantee mechanism, and can reduce network interaction signaling overhead of the terminal in computing unloading and CPU occupancy rate and energy consumption of a terminal resident decision process; the distributed deep learning platform ensures the end-to-end service quality, supports the deep learning model and the customized extension of the algorithm, and can avoid the huge load and energy consumption brought to a wide area network by the traditional mobile cloud computing.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, cloud computing and big data, in particular to an artificial intelligence cloud platform oriented to the 5G era.
Background
Artificial intelligence (ArtificialIntelligence), abbreviated in english as AI. The method is a new technical science for researching and developing theories, methods, technologies and application systems for simulating, extending and expanding human intelligence;
the 5G network is a fifth generation mobile communication network, the peak theoretical transmission speed of which can reach 1Gb per second, which is hundreds of times faster than that of the 4G network, and with the advent of the 5G technology, the era of sharing 3D movies, games, and Ultra High Definition (UHD) programs with intelligent terminals is moving forward;
massive data are generated in the era of interconnection of everything driven by 5G, and the demand on cloud computing is increased; the 5G edge computing can better realize sensing, interaction and control between objects through the data processing capability closer to the application side, and huge incremental space is brought to cloud computing.
When data tsunami type growth, the demand for computing power will be greatly increased, and no good solution and artificial intelligence cloud platform exist for explosive information growth and dynamic flexible architecture demand.
Disclosure of Invention
In order to overcome the defects of the prior art, the embodiment of the invention provides an artificial intelligence cloud platform facing to the 5G era, and the artificial intelligence cloud platform facing to the 5G era is independently developed and developed, a distributed mobile cloud computing cooperative architecture technology is fused, a distributed mobile cloud computing cooperative architecture facing to the 5G era and an end-to-end-terminal computing unloading service quality assurance mechanism system are established, so that the network interaction signaling overhead of a terminal in the computing unloading process and the CPU occupancy rate and energy consumption of a terminal resident decision process can be reduced; the distributed deep learning platform can guarantee the end-to-end service quality of mobile cloud computing, supports the customized extension of a deep learning model and an algorithm, can avoid huge load and energy consumption brought to a wide area network by traditional mobile cloud computing, carries out real-time interaction through a 5G network and a cloud end, improves data processing capacity, reduces time delay, fuses the bottleneck that an AI transmission technology is broken through by a 5G technology, and achieves intelligent enabling.
In order to achieve the purpose, the invention provides the following technical scheme: an artificial intelligence cloud platform facing to the 5G era comprises a heterogeneous distributed cloud platform, wherein the heterogeneous distributed cloud platform takes a distributed mobile cloud computing cooperative architecture as a basic architecture;
the distributed mobile cloud computing cooperative architecture comprises an intelligent terminal, a mobile network, a distributed mobile cloud computing server, a cooperative controller and an end-to-end-terminal computing unloading service quality assurance mechanism system, wherein the distributed mobile cloud computing server is connected with a service base station, and the cooperative controller is connected with a cooperative server;
the intelligent terminal is used as an initiating terminal of mobile cloud computing, periodically uploads self state perception information through a mobile communication network, and receives a collaborative controller computing related unloading segmentation decision;
the mobile network provides wireless access and transmission for the intelligent terminal initiating the calculation unloading request;
the distributed mobile cloud computing server is deployed in a small server or a multi-small server cluster on the mobile access network side, the load state and the virtual machine computing capacity perception information are periodically uploaded to the cooperative server, and the cooperative controller decision information is received to reserve virtual machine resources for computing unloading tasks;
The cooperative controller is used for collecting sensing information of the intelligent terminal, the mobile network and the distributed mobile cloud computing server, generating a computing unloading and dividing decision and issuing the computing unloading and dividing decision to the intelligent terminal, and issuing a resource reservation decision to the service base station and the distributed mobile cloud computing server;
the end-to-end-terminal computing unloading service quality assurance mechanism system comprises a distributed cloud computing sensing module and a collaborative decision module, wherein the distributed cloud computing sensing module works at a server level and a virtual machine level respectively, and the collaborative decision module comprises a mobile terminal part decision information unit, a mobile communication network part decision information unit and a distributed cloud computing node decision information unit;
the heterogeneous distributed cloud platform comprises a heterogeneous distributed artificial intelligence cloud computing center, a distributed deep learning platform, a deep learning large-scale training system, a heterogeneous supercomputing platform and a heterogeneous basic algorithm library;
the deep learning large-scale training system is used for multi-machine multi-CPU-FPGA-GPU hybrid distributed deep learning model training, supports models with billions of parameters and large-scale classification of billions of classes;
the heterogeneous supercomputing platform is used for a plurality of computing clusters, central unified storage and lightweight virtualization and provides continuous computing capability support for researchers;
Various machine learning algorithms including a deep neural network and mathematical and image processing algorithms are stored in the heterogeneous basic algorithm library;
the heterogeneous distributed artificial intelligence cloud computing center is used for realizing artificial intelligence real-time interaction between a 5G network and a cloud end;
the distributed deep learning platform is used for supporting the customized expansion of a deep learning model and an algorithm and supporting the mixed distributed operation of a CPU-GPU or a GPU-GPU or a CPU-FPGA-GPU.
In a preferred embodiment, the cooperative controller is in the form of an instance or virtual machine running in a distributed cloud computing server or coexisting with other network elements of an operator, specifically, a service gateway, a packet data gateway, and a policy and resource management module.
In a preferred embodiment, the intelligent terminal is connected to a mobility management entity through a local gateway, the mobility management entity is connected to a packet data gateway and a policy and resource management device through a serving gateway, the packet data gateway and the policy and resource management device are connected to the resource management device, the packet data gateway and the policy and resource management device are respectively connected to an operator service terminal and the internet, and the intelligent terminal is further connected to the distributed mobile cloud computing server through the local gateway.
In a preferred embodiment, the distributed cloud computing sensing module is configured to collect, at a server level, load conditions of a whole computing node server or a server cluster, specifically including server throughput, server concurrent communication status, server computing resource usage, server storage resource occupancy, and when a computing node is a server cluster and a virtualization technology is used in the cluster to implement a virtualized resource pool of the whole cluster, the server level sensing information should fully consider the whole conditions of the virtual resource pool.
In a preferred embodiment, the distributed cloud computing awareness module is used at the virtual machine level to collect virtual machine state information in the whole node, where the information includes: the number of virtual machines, the amount of computing and storage resources occupied by each virtual machine, the equivalent throughput generated by each virtual machine, and the related state information of the bearing resources.
In a preferred embodiment, the cooperative controller fully grasps the state information of the intelligent terminal, the mobile network and each distributed cloud computing node by acquiring the perception information of the intelligent terminal, the mobile communication network and the distributed mobile cloud computing server, the cooperative decision module generates a cooperative decision by comprehensively analyzing the grasped information and issues the cooperative decision to the intelligent terminal, the mobile network element and each distributed cloud computing node, and each part executes the corresponding action according to the decision.
In a preferred embodiment, the mobile terminal part decision information unit is configured to determine calculation subtask division of a corresponding mobile application according to a current battery, energy consumption, and calculation resource state of the intelligent terminal, and further plan a local calculation task and an offload calculation task according to a wireless bandwidth resource state of the intelligent terminal, where when the terminal accesses multiple base stations, a corresponding access base station is assigned to each offload calculation subtask, and at this time, the mobile terminal part decision information unit correctly reassembles calculation subtask results returned out of order.
In a preferred embodiment, the mobile communication network part decision information unit is configured to allocate corresponding access points to offload computation task data and return result data to complete receiving and sending according to a wireless bandwidth state between the intelligent terminal and each access base station and a backhaul network congestion state of each base station.
In a preferred embodiment, the distributed cloud computing node decision information unit is configured to assign a corresponding distributed cloud computing node to each offload computing sub-task according to state information of each distributed cloud computing node to complete a computing load-bearing task; the system comprises a plurality of distributed cloud computing nodes, a plurality of virtual machines and a plurality of virtual machine configuration modules, wherein the distributed cloud computing nodes are used for deciding that each distributed cloud computing node bears the virtualized resources required by a corresponding unloading computing subtask, and a virtual machine is generated to complete a computing task; the method is used for realizing parallel cooperation of computing processes among a plurality of distributed cloud computing nodes.
The invention has the technical effects and advantages that:
1. aiming at the requirements of enhancing a mobile broadband, being low in time delay, high in reliability, large in connection and low in power consumption in an application scene of an AI cloud computing service system in the 5G era, the technical framework defects of the existing network cloud computing service platform are improved, a high-performance heterogeneous distributed cloud platform facing to the 5G era is independently developed, a distributed mobile cloud computing cooperative framework technology is fused, a 5G-facing distributed mobile cloud computing cooperative framework and an end-to-end-terminal computing unloading service quality guarantee mechanism system are established, and network interaction signaling overhead of a terminal in computing unloading and CPU occupancy rate and energy consumption of a terminal resident decision process can be reduced; the distributed deep learning platform can guarantee the end-to-end service quality of mobile cloud computing, support the deep learning model and the customized extension of the algorithm, and avoid the huge load and energy consumption brought to a wide area network by the traditional mobile cloud computing;
2. in the 5G era, a high-performance heterogeneous distributed cloud platform is adopted, a deep learning large-scale training system is combined, a strong enterprise artificial intelligence innovation service solution facing the 5G intelligent era is provided for a user by utilizing a heterogeneous high-performance computing center and a high-performance heterogeneous basic algorithm library, the characteristics of high speed and low time delay of 5G are fully utilized, intelligent cooperation of a cloud AI and a terminal-side AI is promoted, real-time interaction is carried out through a 5G network and a cloud, the data processing capacity is improved, the time delay is reduced, the bottleneck of an AI transmission technology is broken through by fusing the 5G technology, and intelligent enabling is realized;
3. The terminal side AI can quickly respond to the user requirements, quickly display the processed image, video, voice and text information to the user in a low-power consumption and low-cost mode, and is suitable for finishing an AI reasoning task; the integration of the 5G technology enables intelligent collaborative innovative integration of the cloud AI and the terminal side AI to be possible, real-time interaction is carried out through the 5G network and the cloud, data processing capacity is improved, and time delay is reduced; the cloud AI is used for realizing multi-terminal data aggregation, has more advantages in the aspects of data throughput, processing speed and the like, and is suitable for finishing large-scale large-data-volume AI model training tasks.
Drawings
Fig. 1 is a block diagram of the overall structure of the present invention.
Fig. 2 is a structural block diagram of a distributed mobile cloud computing collaboration architecture of the present invention.
Fig. 3 is a topological diagram of a distributed mobile cloud computing collaborative architecture according to the present invention.
Fig. 4 is a system topology diagram of an end-to-end computation offload quality of service assurance mechanism of the present invention.
FIG. 5 is a structural diagram of a distributed deep learning platform according to the present invention.
FIG. 6 is a schematic structural diagram of a deep learning large-scale training system according to the present invention.
FIG. 7 is a schematic diagram of a heterogeneous supercomputing platform according to the present invention.
FIG. 8 is a diagram illustrating a heterogeneous basic algorithm library structure according to the present invention.
The reference signs are: the system comprises a heterogeneous distributed cloud platform 1, a heterogeneous distributed artificial intelligence cloud computing center 11, a distributed deep learning platform 12, a deep learning large-scale training system 13, a heterogeneous supercomputing platform 14, a heterogeneous basic algorithm library 15, a distributed mobile cloud computing cooperative framework 2, an intelligent terminal 21, a mobile network 22, a distributed mobile cloud computing server 23, a cooperative controller 24, an end-to-end-terminal computing unloading service quality guarantee mechanism system 25 and a service base station 3.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The artificial intelligence cloud platform facing the 5G era shown in fig. 1-3 comprises a heterogeneous distributed cloud platform 1, wherein the heterogeneous distributed cloud platform 1 takes a distributed mobile cloud computing cooperative architecture 2 as a basic architecture;
The mobile cloud computing unloads the terminal computing task to the cloud, so that the mobile application capability of realizing large computing amount under the condition of limited terminal resources is provided while the energy consumption of the terminal is reduced, and the mobile cloud computing is bound to become an essential key technology for constructing future mobile internet innovation services; however, in the current-stage technology development of mobile cloud computing, an effective countermeasure is still lacking for problems of end-to-end network delay and bandwidth instability from mobile equipment to a cloud computing center, dynamics of main factors influencing the quality of mobile cloud computing service and the like, and an effective mechanism design is lacking in the aspects of terminal perception and negotiation decision in computing unloading, so that the distributed mobile cloud computing cooperative architecture 2 is provided by combining the characteristics of 4G and 5G network wireless access network technologies;
the distributed mobile cloud computing collaborative architecture 2 comprises an intelligent terminal 21, a mobile network 22, a distributed mobile cloud computing server 23, a collaborative controller 24 and an end-to-end-terminal computing unloading service quality assurance mechanism system 25, wherein the distributed mobile cloud computing server 23 is connected with the service base station 3, and the collaborative controller 24 is connected with the collaborative server;
the intelligent terminal 21 is used as an initiating terminal of mobile cloud computing, periodically uploads self state perception information through a mobile communication network, and receives a related unloading and dividing decision calculated by the cooperative controller 24;
The mobile network 22 provides wireless access and transmission for the intelligent terminal 21 which initiates the calculation unloading request;
the distributed mobile cloud computing server 23 is deployed in a small server or a cluster of multiple small servers on the mobile access network side, the load state and the virtual machine computing capacity perception information of the distributed mobile cloud computing server are periodically uploaded to the cooperative server, and the cooperative controller 24 decision information is received to reserve virtual machine resources for computing unloading tasks;
the cooperative controller 24 is configured to collect the sensing information of the intelligent terminal 21, the mobile network 22 and the distributed mobile cloud computing server 23, generate a computation offload partition decision and send the computation offload partition decision to the intelligent terminal 21, and send a resource reservation decision to the service base station 3 and the distributed mobile cloud computing server 23;
the form of the cooperative controller 24 is an instance or a virtual machine running in a distributed cloud computing server or coexisting in other network elements of an operator, specifically, a service gateway, a packet data gateway, and a policy and resource management module;
the intelligent terminal 21 is connected with a mobility management entity through a local gateway, the mobility management entity is connected with a packet data gateway and a policy and resource management device through a service gateway, the packet data gateway and the policy and resource management device are connected with an operator service terminal and the internet respectively, and the intelligent terminal 21 is also connected with a distributed mobile cloud computing server 23 through the local gateway;
The implementation mode is specifically as follows: by fusing the technology of the distributed mobile cloud computing cooperative architecture 2, the 5G-oriented distributed mobile cloud computing cooperative architecture 2 and the end-to-end-terminal computing unloading service quality assurance mechanism system 25 are established, so that the network interaction signaling overhead of the terminal in the computing unloading process and the CPU occupancy rate and energy consumption of the terminal resident decision process can be reduced; the end-to-end service quality of the mobile cloud computing can be ensured; huge load and energy consumption brought to WAN by traditional mobile cloud computing can be avoided,
it is expected that under the promotion of multiple parties of users, intelligent equipment vendors and mobile network 22 operators, distributed mobile cloud computing will become one of mainstream technologies of 4G and 5G networks in the future, become an infrastructure of mobile operators and provide new service growth points for the mobile operators, and provide better user experience for the application of the intelligent terminal 21.
Although the technology of the 5G mobile network 22 is still in the process of preliminary research, many potential technologies foreseen by the 5G network provide effective feasible support for the distributed mobile cloud computing proposed above, and the potential technologies are as follows:
a. ultra-dense bee deployment
The idea of LTE4G network Femtocell deployment, such as a femto base station (Femtocell) and a pico base station (Picocell), is continued to improve the network coverage quality and the network capacity. The deployment of the 5G network micro base stations is more intensive, and under the condition that the problems of same frequency interference and wireless resource reuse are effectively solved, the wireless networking mode can bear more mobile network 22 transmissions, and the energy consumption and time delay of wireless data transmission of the intelligent terminal 21 in mobile cloud computing can be effectively reduced due to the fact that the average wireless transmission distance between the access point and the intelligent terminal 21 is shortened.
b. Massive MIMO
As a derivative enhancement of a multiple-input multiple-output (MIMO) technology, a large-scale MIMO (masivemimo) deploys a larger number (tens or even hundreds) of antenna units at a transmitting end and/or a receiving end, so that a large increase of a space beamforming gain can be achieved through a simple linear precoding algorithm, and communication reliability, link throughput, spectrum efficiency and energy efficiency of a point-to-point/multipoint link are significantly enhanced, so that more concurrent mobile cloud computing transmissions can be carried at a high rate.
c. Millimeter wave backhaul
With the rapid increase of the number of base stations, a backhaul network connecting the macro base station, the micro base station and the mobile switching node becomes a key for ensuring the network performance. Currently, millimeter waves (including frequencies of 71-76, 81-86, 92-95 GHz, etc.) have been used as a carrier for 4G backhaul networks, and it is necessary to further enhance the carrier in 5G networks. The microwave backhaul provides guarantee for backhaul bandwidth and time delay of a macro base station and a micro base station in a 5G network, and particularly can effectively solve the problems of time delay jitter and congestion of an ADSL backhaul link of the micro base station, so that sensing uploading of the intelligent terminal 21 and the base station to the cooperative controller 24 in the distributed mobile cloud computing cooperative structure and decision issuing of the cooperative controller 24 to the terminal and the base station are supported at a low time delay and a high speed.
As shown in fig. 4, the computation task unloaded by the terminal through the mobile cloud computing is finally completed in the virtual machines distributed in the distributed cloud computing servers in the mobile access network, so that distributed cloud computing perception including the load of each distributed computing node, the number of virtual machines that can be borne in each node, and the computing power of each virtual machine is an important basis for selecting the computing unloading node, and has an important influence on the mobile cloud computing performance;
from the above, the end-to-end-terminal computing offload service quality assurance mechanism system 25 includes a distributed cloud computing sensing module and a collaborative decision module, the distributed cloud computing sensing module respectively performs work at a server level and a virtual machine level, and the collaborative decision module includes a mobile terminal part decision information unit, a mobile communication network part decision information unit and a distributed cloud computing node decision information unit;
the distributed cloud computing perception module is used for collecting the load condition of the whole computing node server or server cluster at the server level, and specifically comprises server throughput, server concurrent communication state, server computing resource use condition, server storage resource occupancy rate and the whole condition of a virtual resource pool when the computing node is the server cluster and the virtualization technology is adopted in the cluster to realize the virtual resource pool of the whole cluster, wherein the server level perception information fully considers the whole condition of the virtual resource pool;
The distributed cloud computing perception module is used for collecting the state information of the virtual machine in the whole node at the virtual machine level, and the information comprises: the number of the virtual machines, the amount of computing and storage resources occupied by each virtual machine, the equivalent throughput generated by each virtual machine and the related state information of the bearing resources;
at present, mainstream server products support software acquisition and storage of the information, and distributed cloud computing high-efficiency perception of a cooperative controller is realized through a high-efficiency interface and an information sharing mechanism;
the cooperative controller 24 fully grasps the state information of the intelligent terminal 21, the mobile network 22 (including an access network and a core network) and each distributed cloud computing node by acquiring the perception information of the intelligent terminal 21, the mobile communication network and the distributed mobile cloud computing server 23, the cooperative decision module generates a cooperative decision by comprehensively analyzing the grasped information and respectively issues the cooperative decision to the intelligent terminal 21, the mobile network 22 network element and the distributed cloud computing node, each part executes respective corresponding action according to the decision to realize cooperative distributed cloud computing, and user experience guarantee is provided for terminal application;
the mobile terminal part decision information unit is used for determining the division of calculation subtasks of corresponding mobile applications according to the current battery, energy consumption and calculation resource states of the intelligent terminal 21, and further planning a local calculation task and an unloading calculation task according to the wireless bandwidth resource states of the intelligent terminal 21;
The mobile communication network part decision information unit is used for distributing corresponding access points for unloading calculation task data and returning result data to complete receiving and sending according to the wireless bandwidth state between the intelligent terminal 21 and each access base station and the return network congestion state of each base station, and specifically comprises the following steps:
when the terminal application wants to initiate a computation uninstalling request at a far end, informing a corresponding base station to receive air interface data containing a corresponding far end computation subtask from the terminal;
when the distributed cloud computing node returns the sub-computing task result, assigning the corresponding base station to send air interface data containing the returned sub-computing task result to the designated terminal;
considering user mobility, the sending and receiving base stations of the request and result of the same sub-computing task need to be dynamically assigned respectively; when the offloaded computing task data reaches the core network of the mobile network 22, routing is selected for each offloaded computing subtask data according to the network state (topology, link flow and node load) borne by the core network, so that efficient forwarding of each offloaded computing subtask data to a designated distributed cloud computing node is realized, and the network load and time delay of the core network are reduced;
otherwise, when the distributed cloud computing nodes return the unloading computing sub-task results, the distributed cloud computing nodes provide routing forwarding to the assigned access network nodes; the process can realize the dynamic adaptation of the routing forwarding plane by depending on a software defined network technology;
The distributed cloud computing node decision information unit is used for assigning corresponding distributed cloud computing nodes to each unloading computing sub task according to the state information of each distributed cloud computing node to complete a computing bearing task; the system comprises a plurality of distributed cloud computing nodes, a plurality of virtual machines and a plurality of virtual machine configuration modules, wherein the distributed cloud computing nodes are used for deciding that each distributed cloud computing node bears the virtualized resources required by a corresponding unloading computing subtask, and a virtual machine is generated to complete a computing task; the method is used for realizing parallel cooperation of computing processes among a plurality of distributed cloud computing nodes.
As shown in fig. 5-8, the heterogeneous distributed cloud platform 1 includes a heterogeneous distributed artificial intelligence cloud computing center 11, a distributed deep learning platform 12, a deep learning large-scale training system 13, a heterogeneous supercomputing platform 14, and a heterogeneous basic algorithm library 15;
the heterogeneous distributed artificial intelligence cloud computing center 11 is used for realizing artificial intelligence real-time interaction between a 5G network and a cloud end; the distributed deep learning platform 12 is used for supporting the customized expansion of a deep learning model and an algorithm and supporting the CPU-GPU or GPU-GPU or CPU-FPGA-GPU hybrid distributed operation.
The deep learning large-scale training system 13 is used for multi-machine multi-CPU-FPGA-GPU hybrid distributed deep learning model training, a model supporting billions of parameters, large-scale classification of billions of categories, an industry-leading memory optimization and communication optimization technology, and hundreds of CPU-FPGA-GPU hybrid distributed combined training, so that the speed of company training and iterative model is greatly improved, and is shown in FIG. 6;
The heterogeneous supercomputing platform 14 is used for a plurality of computing clusters, central unified storage, lightweight virtualization and support for providing continuous computing power for researchers, as shown in fig. 7;
the heterogeneous basic algorithm library 15 stores various machine learning algorithms including a deep neural network and mathematics and image processing algorithms; compared with an open source platform library in the industry, the performance is improved by 2-5 times. The method supports mainstream cloud, personal computer, mobile terminal and embedded terminal hardware platforms, and supports various system platforms such as Linux, Android, iOS and Windows, and is shown in fig. 8.
The implementation mode is specifically as follows: the heterogeneous distributed cloud platform 1 supports the distributed deep learning platform 12 with the deep learning model and the algorithm customized expansion, and can avoid huge load and energy consumption brought to a Wide Area Network (WAN) by traditional mobile cloud computing;
the integration of the 5G technology is realized through a heterogeneous distributed artificial intelligence cloud platform computing center, so that intelligent collaborative innovation and integration of the far-end AI and the terminal-side AI become possible, real-time interaction is performed through a 5G network and a cloud, the data processing capacity is improved, the time delay is reduced, the bottleneck of an AI transmission technology is broken through by integrating the 5G technology, and intelligent enabling is realized;
In the 5G era, the invention adopts the high-performance heterogeneous distributed cloud platform 1, combines a deep learning large-scale training system 13, and provides a strong enterprise artificial intelligence innovation service solution facing the 5G intelligent era for users by utilizing a heterogeneous high-performance computing center and a high-performance heterogeneous basic algorithm library 15.
The points to be finally explained are: first, in the description of the present application, it should be noted that, unless otherwise specified and limited, the terms "mounted," "connected," and "connected" should be understood broadly, and may be a mechanical connection or an electrical connection, or a communication between two elements, and may be a direct connection, and "upper," "lower," "left," and "right" are only used to indicate a relative positional relationship, and when the absolute position of the object to be described is changed, the relative positional relationship may be changed;
secondly, the method comprises the following steps: in the drawings of the disclosed embodiments of the invention, only the structures related to the disclosed embodiments are referred to, other structures can refer to common designs, and the same embodiment and different embodiments of the invention can be combined with each other without conflict;
and finally: the above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that are within the spirit and principle of the present invention are intended to be included in the scope of the present invention.
Claims (9)
1. The utility model provides an artificial intelligence cloud platform towards 5G era which characterized in that: the heterogeneous distributed cloud platform system comprises a heterogeneous distributed cloud platform (1), wherein the heterogeneous distributed cloud platform (1) takes a distributed mobile cloud computing cooperative architecture (2) as a basic architecture;
the distributed mobile cloud computing collaborative architecture (2) comprises an intelligent terminal (21), a mobile network (22), a distributed mobile cloud computing server (23), a collaborative controller (24) and an end-to-end-terminal computing unloading service quality assurance mechanism system (25), wherein the distributed mobile cloud computing server (23) is connected with a service base station (3), and the collaborative controller (24) is connected with the collaborative server;
the intelligent terminal (21) is used as an initiating terminal of mobile cloud computing, uploads self state perception information periodically through a mobile communication network, and receives a related unloading and partitioning decision calculated by the cooperative controller (24);
the mobile network (22) provides wireless access and transmission for the intelligent terminal (21) which initiates the calculation unloading request;
the distributed mobile cloud computing server (23) is deployed in a small server or a multi-small server cluster on the mobile access network side, the load state and the virtual machine computing capacity perception information are periodically uploaded to the cooperative server, and the decision information of the cooperative controller (24) is received to reserve virtual machine resources for computing unloading tasks;
The cooperative controller (24) is used for collecting perception information of the intelligent terminal (21), the mobile network (22) and the distributed mobile cloud computing server (23), generating a computation unloading segmentation decision and issuing the computation unloading segmentation decision to the intelligent terminal (21), and issuing a resource reservation decision to the service base station (3) and the distributed mobile cloud computing server (23);
the end-to-end-terminal computing unloading service quality assurance mechanism system (25) comprises a distributed cloud computing sensing module and a cooperative decision module, wherein the distributed cloud computing sensing module works at a server level and a virtual machine level respectively, and the cooperative decision module comprises a mobile terminal part decision information unit, a mobile communication network part decision information unit and a distributed cloud computing node decision information unit;
the heterogeneous distributed cloud platform (1) comprises a heterogeneous distributed artificial intelligence cloud computing center (11), a distributed deep learning platform (12), a deep learning large-scale training system (13), a heterogeneous super computing platform (14) and a heterogeneous basic algorithm library (15);
the deep learning large-scale training system (13) is used for multi-machine multi-GPU distributed deep learning model training, supports models with billions of parameters and large-scale classification of billions of classes;
The heterogeneous supercomputing platform (14) is used for a plurality of computing clusters, central unified storage, lightweight virtualization and provides continuous computing capability support for researchers;
the heterogeneous basic algorithm library (15) stores various machine learning algorithms including a deep neural network and mathematics and image processing algorithms;
the heterogeneous distributed artificial intelligence cloud computing center (11) is used for realizing artificial intelligence real-time interaction between a 5G network and a cloud end;
the distributed deep learning platform (12) is used for supporting the customized extension of a deep learning model and an algorithm and supporting a large number of general CPUs and GPUs or the mixed distributed operation of the CPUs and the GPUs.
2. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the cooperative controller (24) is in the form of an instance or virtual machine running in a distributed cloud computing server or coexisting in other network elements of an operator, and specifically comprises a service gateway, a packet data gateway and a policy and resource management module.
3. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the intelligent terminal (21) is connected with a mobility management entity through a local gateway, the mobility management entity is connected with a packet data gateway and a strategy and resource management device through a service gateway, the packet data gateway and the strategy and resource management device are connected with an operator service terminal and the internet respectively, and the intelligent terminal (21) is also connected with a distributed mobile cloud computing server (23) through the local gateway.
4. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the distributed cloud computing perception module is used for collecting the load condition of the whole computing node server or a server cluster at a server level, and specifically comprises server throughput, a server concurrent communication state, a server computing resource use condition, a server storage resource occupancy rate and the whole condition of a virtual resource pool when the computing node is the server cluster and the virtualization technology is adopted in the cluster to realize the virtual resource pool of the whole cluster, wherein the server level perception information fully considers the whole condition of the virtual resource pool.
5. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the distributed cloud computing perception module is used for collecting the state information of the virtual machine in the whole node at the virtual machine level, and the information comprises: the number of virtual machines, the amount of computing and storage resources occupied by each virtual machine, the equivalent throughput generated by each virtual machine, and the related state information of the bearing resources.
6. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the cooperative controller (24) fully grasps the state information of the intelligent terminal (21), the mobile network (22) and each distributed cloud computing node by acquiring the perception information of the intelligent terminal (21), the mobile communication network and the distributed mobile cloud computing server (23), the cooperative decision module generates a cooperative decision by comprehensively analyzing the grasped information and respectively issues the cooperative decision to the intelligent terminal (21), the mobile network (22) network element and the distributed cloud computing node, and each part executes the corresponding action according to the decision.
7. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the mobile terminal part decision information unit is used for determining the division of calculation subtasks of corresponding mobile applications according to the current battery, energy consumption and calculation resource states of the intelligent terminal (21), and further planning a local calculation task and an unloading calculation task according to the wireless bandwidth resource states of the intelligent terminal (21).
8. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the mobile communication network part decision information unit is used for distributing corresponding access points for unloading calculation task data and returning result data to complete receiving and sending according to the wireless bandwidth state between the intelligent terminal (21) and each access base station and the return network congestion state of each base station.
9. The artificial intelligence cloud platform for the age of 5G according to claim 1, wherein: the distributed cloud computing node decision information unit is used for assigning corresponding distributed cloud computing nodes to each unloading computing sub task according to the state information of each distributed cloud computing node to complete a computing bearing task; the system comprises a plurality of distributed cloud computing nodes, a plurality of virtual machines and a plurality of virtual machine configuration modules, wherein the distributed cloud computing nodes are used for deciding that each distributed cloud computing node bears the virtualized resources required by a corresponding unloading computing subtask, and a virtual machine is generated to complete a computing task; the method is used for realizing parallel cooperation of computing processes among a plurality of distributed cloud computing nodes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910426197.8A CN111984364B (en) | 2019-05-21 | 2019-05-21 | Artificial intelligence cloud platform towards 5G age |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910426197.8A CN111984364B (en) | 2019-05-21 | 2019-05-21 | Artificial intelligence cloud platform towards 5G age |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111984364A true CN111984364A (en) | 2020-11-24 |
| CN111984364B CN111984364B (en) | 2023-05-26 |
Family
ID=73436212
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910426197.8A Active CN111984364B (en) | 2019-05-21 | 2019-05-21 | Artificial intelligence cloud platform towards 5G age |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111984364B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112512069A (en) * | 2021-02-02 | 2021-03-16 | 网络通信与安全紫金山实验室 | Network intelligent optimization method and device based on channel beam pattern |
| CN112637290A (en) * | 2020-12-14 | 2021-04-09 | 厦门宏泰科技研究院有限公司 | Global communication network system based on micro base station and edge calculation |
| CN112804446A (en) * | 2020-12-30 | 2021-05-14 | 江苏达科信息科技有限公司 | Big data processing method and device based on cloud platform big data |
| CN113067873A (en) * | 2021-03-19 | 2021-07-02 | 北京邮电大学 | Edge cloud collaborative optimization method based on deep reinforcement learning |
| CN113222170A (en) * | 2021-03-30 | 2021-08-06 | 新睿信智能物联研究院(南京)有限公司 | Intelligent algorithm and model for IOT (Internet of things) AI (Artificial Intelligence) collaborative service platform |
| CN113301141A (en) * | 2021-05-20 | 2021-08-24 | 北京邮电大学 | Deployment method and system of artificial intelligence support framework |
| CN113316159A (en) * | 2021-05-27 | 2021-08-27 | 北京邮电大学 | 5G network system based on heterogeneous physical layer |
| CN113473449A (en) * | 2021-07-02 | 2021-10-01 | 深圳市威宇智通科技有限公司 | Intelligent connection system based on Internet of things terminal |
| CN115002038A (en) * | 2022-08-04 | 2022-09-02 | 中国电子科技集团公司第十五研究所 | Intelligent peak regulation method and system based on cloud distributed coordination service |
| CN115442202A (en) * | 2022-08-15 | 2022-12-06 | 鞍钢集团北京研究院有限公司 | 5G network cloud-end cooperative virtual PLC control method and system |
| WO2023098665A1 (en) * | 2021-11-30 | 2023-06-08 | 华为技术有限公司 | Task processing method and apparatus |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102681899A (en) * | 2011-03-14 | 2012-09-19 | 金剑 | Virtual computing resource dynamic management system of cloud computing service platform |
| US20160283284A1 (en) * | 2015-03-27 | 2016-09-29 | Mingqiu Sun | Technologies for offloading and on-loading data for processor/coprocessor arrangements |
| US20170230448A1 (en) * | 2016-02-05 | 2017-08-10 | International Business Machines Corporation | Context-aware task offloading among multiple devices |
| US20180349168A1 (en) * | 2017-05-30 | 2018-12-06 | Magalix Corporation | Systems and methods for managing a cloud computing environment |
| US20190095245A1 (en) * | 2017-09-22 | 2019-03-28 | CloudHealth Technologies, Inc. | System and Method for Apportioning Shared Computer Resources |
| US20190121566A1 (en) * | 2017-10-19 | 2019-04-25 | Pure Storage, Inc. | Data transformation offloading in an artificial intelligence infrastructure |
-
2019
- 2019-05-21 CN CN201910426197.8A patent/CN111984364B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102681899A (en) * | 2011-03-14 | 2012-09-19 | 金剑 | Virtual computing resource dynamic management system of cloud computing service platform |
| US20160283284A1 (en) * | 2015-03-27 | 2016-09-29 | Mingqiu Sun | Technologies for offloading and on-loading data for processor/coprocessor arrangements |
| US20170230448A1 (en) * | 2016-02-05 | 2017-08-10 | International Business Machines Corporation | Context-aware task offloading among multiple devices |
| US20180349168A1 (en) * | 2017-05-30 | 2018-12-06 | Magalix Corporation | Systems and methods for managing a cloud computing environment |
| US20190095245A1 (en) * | 2017-09-22 | 2019-03-28 | CloudHealth Technologies, Inc. | System and Method for Apportioning Shared Computer Resources |
| US20190121566A1 (en) * | 2017-10-19 | 2019-04-25 | Pure Storage, Inc. | Data transformation offloading in an artificial intelligence infrastructure |
Non-Patent Citations (4)
| Title |
|---|
| CHOY SHARON ET AL.: "A hybrid edge-cloud architecture for reducing on-demand gaming latency", MULTIMEDIA SYSTEMS * |
| SAIYEDA, ANAM ET AL.: "Cloud computing for deep learning analytics: A survey of current trends and challenges", INTERNATIONAL JOURNAL OF ADVANCED RESEARCH IN COMPUTER SCIENCE * |
| 廉晓飞等: "基于安全保障的边缘计算卸载方案", 中兴通讯技术 * |
| 李宏佳;陈鑫;周旭;: "面向55G的分布式移动云计算协同架构与管理机制", 中兴通讯技术 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112637290A (en) * | 2020-12-14 | 2021-04-09 | 厦门宏泰科技研究院有限公司 | Global communication network system based on micro base station and edge calculation |
| CN112637290B (en) * | 2020-12-14 | 2024-03-19 | 厦门宏泰科技研究院有限公司 | Global communication network system based on micro base station and edge calculation |
| CN112804446B (en) * | 2020-12-30 | 2022-08-12 | 江苏达科信息科技有限公司 | Big data processing method and device based on cloud platform big data |
| CN112804446A (en) * | 2020-12-30 | 2021-05-14 | 江苏达科信息科技有限公司 | Big data processing method and device based on cloud platform big data |
| CN112512069B (en) * | 2021-02-02 | 2021-05-28 | 网络通信与安全紫金山实验室 | Network intelligent optimization method and device based on channel beam pattern |
| CN112512069A (en) * | 2021-02-02 | 2021-03-16 | 网络通信与安全紫金山实验室 | Network intelligent optimization method and device based on channel beam pattern |
| CN113067873A (en) * | 2021-03-19 | 2021-07-02 | 北京邮电大学 | Edge cloud collaborative optimization method based on deep reinforcement learning |
| CN113222170A (en) * | 2021-03-30 | 2021-08-06 | 新睿信智能物联研究院(南京)有限公司 | Intelligent algorithm and model for IOT (Internet of things) AI (Artificial Intelligence) collaborative service platform |
| CN113222170B (en) * | 2021-03-30 | 2024-04-23 | 新睿信智能物联研究院(南京)有限公司 | Intelligent algorithm and model for AI collaborative service platform of Internet of things |
| CN113301141A (en) * | 2021-05-20 | 2021-08-24 | 北京邮电大学 | Deployment method and system of artificial intelligence support framework |
| CN113301141B (en) * | 2021-05-20 | 2022-06-17 | 北京邮电大学 | Deployment method and system of artificial intelligence support framework |
| CN113316159B (en) * | 2021-05-27 | 2022-09-02 | 北京邮电大学 | 5G network system based on heterogeneous physical layer |
| CN113316159A (en) * | 2021-05-27 | 2021-08-27 | 北京邮电大学 | 5G network system based on heterogeneous physical layer |
| CN113473449A (en) * | 2021-07-02 | 2021-10-01 | 深圳市威宇智通科技有限公司 | Intelligent connection system based on Internet of things terminal |
| WO2023098665A1 (en) * | 2021-11-30 | 2023-06-08 | 华为技术有限公司 | Task processing method and apparatus |
| CN115002038A (en) * | 2022-08-04 | 2022-09-02 | 中国电子科技集团公司第十五研究所 | Intelligent peak regulation method and system based on cloud distributed coordination service |
| CN115442202A (en) * | 2022-08-15 | 2022-12-06 | 鞍钢集团北京研究院有限公司 | 5G network cloud-end cooperative virtual PLC control method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111984364B (en) | 2023-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111984364B (en) | Artificial intelligence cloud platform towards 5G age | |
| Qiao et al. | Collaborative task offloading in vehicular edge multi-access networks | |
| Islam et al. | A survey on task offloading in multi-access edge computing | |
| Xiao et al. | Distributed optimization for energy-efficient fog computing in the tactile internet | |
| Baccarelli et al. | Fog of everything: Energy-efficient networked computing architectures, research challenges, and a case study | |
| Xiao et al. | Dynamic network slicing for scalable fog computing systems with energy harvesting | |
| Zhou et al. | Machine learning-based offloading strategy for lightweight user mobile edge computing tasks | |
| Pooranian et al. | A novel distributed fog-based networked architecture to preserve energy in fog data centers | |
| CN112333717A (en) | 5G access network slice resource allocation method and device considering power multi-service requirements | |
| Maiti et al. | An effective approach of latency-aware fog smart gateways deployment for IoT services | |
| Li et al. | K-means based edge server deployment algorithm for edge computing environments | |
| Meng et al. | Integration application of 5g and smart grid | |
| Haitao et al. | Multipath transmission workload balancing optimization scheme based on mobile edge computing in vehicular heterogeneous network | |
| Wang et al. | Task allocation mechanism of power internet of things based on cooperative edge computing | |
| Ku et al. | Sustainable vehicular edge computing using local and solar-powered roadside unit resources | |
| Sun et al. | Harmonizing artificial intelligence with radio access networks: Advances, case study, and open issues | |
| Kuo et al. | An integrated edge and fog system for future communication networks | |
| Shukla et al. | Software-defined network based resource allocation in distributed servers for unmanned aerial vehicles | |
| Chen et al. | An IoT edge computing system architecture and its application | |
| Badshah et al. | Use of regional computing to minimize the social big data effects | |
| Kavyashree et al. | Survey on computation offloading strategies in cellular networks with mobile edge computing | |
| CN115955711A (en) | Air-ground 6G network resource allocation method oriented to optimal energy efficiency | |
| Al-Razgan et al. | A computational offloading method for edge server computing and resource allocation management | |
| Yin et al. | An optimization method for resource allocation in fog computing | |
| Liu et al. | Connectivity enhancement of edge computing over fiber-wireless access networks for IoT |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |