CN109828831A - A kind of artificial intelligence cloud platform - Google Patents
A kind of artificial intelligence cloud platform Download PDFInfo
- Publication number
- CN109828831A CN109828831A CN201910112462.5A CN201910112462A CN109828831A CN 109828831 A CN109828831 A CN 109828831A CN 201910112462 A CN201910112462 A CN 201910112462A CN 109828831 A CN109828831 A CN 109828831A
- Authority
- CN
- China
- Prior art keywords
- kloud
- artificial intelligence
- abstracted
- cloud platform
- level
- 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
Landscapes
- Stored Programmes (AREA)
Abstract
The invention discloses a kind of artificial intelligence cloud platforms, successively it is made of IaaS, PaaS and SaaS three-tier architecture, it is successively abstracted using to resource, component and application, the mode thoroughly decoupled to system constructs this cloud platform, IaaS layers use virtualization technology, unified distribution and management bottom hardware resource, the infrastructure for constructing intelligent cloud platform service;PaaS layers by carrying out the abstract autonomous Row control engine of building of level Four to artificial intelligence application scene, and then artificial intelligence component and system service are constructed, it includes following four parts that it is abstract, which to carry out level Four, to artificial intelligence application scene: data level is abstracted KLoud.cube, functional level is abstracted KLoud.case, service level is abstracted KLoud.chain and optimization level is abstracted KLoud.cycle;SaaS layer building artificial intelligence application service.The present invention is based on autonomous Row control engine, it can be achieved that many flexibly operate effectively, solves stacking open source engine and increase system complexity, the disadvantage all very unfavorable to deployment, debugging and maintenance, this cloud platform system has high resiliency and high flexibility.
Description
Technical field
The invention belongs to cloud platform technical fields, and in particular to a kind of artificial intelligence cloud platform.
Background technique
AI intelligent use is also being rapidly developed with the development of society.AI intelligent use is broadly divided into conventional monomer at present
Using, common AI platform and AI cloud platform.
Conventional monomer, which is applied, to be released comprising seen most intelligent security guards application and Haikang, big China etc. on the market
Edge calculations product.Such application does not have ability of second development substantially, is unable to satisfy user and customizes demand.Monomer is answered
For, equipment damage then means that service is whole unavailable, and the period for repairing replacement is long and at high cost.Furthermore monomer applications base
This does not have elasticity, therefore its peak load situation must be taken into consideration in its hardware performance, has in general use process greatly
Cost waste.
Common AI platform such as " Baidu's AI open platform (ai.***.com) ", " spacious view artificial intelligence open platform
(www.faceplusplus.com.cn) " etc., such platform generally externally provides service with SDK and API two ways.For
The difficulty of the method for service of SDK, secondary development is higher, and general integrator and fan are difficult to deal with its technical threshold;For
The method of service of API then requires deployed environment that must have public network to access and meets certain delay requirement, and many scenes
Deployed environment is unable to satisfy above-mentioned requirements.Meanwhile which kind of either above-mentioned method of service, the interface type provided have very much
Limit, such as face alignment service, only provides 1: 1 and 1: N two ways inputs picture and returns to alignment similarity, it is difficult to adapt to as strange
The application scenarios such as people's identification, can generate great computing resource waste.
There is also some other on the market at present compared with the AI cloud platform of minority, such as according to figure, seven Ns of clouds, for specific big
AI intellectual analysis under data scene (such as public security, traffic), a portion use the architecture designs such as k8s, micro services.Mesh
These preceding AI cloud platforms are applied compared to conventional monomer and common AI platform is to be more suitable diversified artificial intelligence application now
Requirement.But such AI cloud platform is using directly stacking such as MapReduce/Spark/Storm/Flink/
The open source big data batch processing/stream process engine such as SparkStream, such mode have following deficiency: (1) from O&M angle
Consider, stack open source engine and increase system complexity, for deployment, debugging and safeguards all very unfavorable, some O&Ms, test
Means such as " canary test " can not be put to good use, and after iteration is applied multiple times, the processing logical gate for engine of increasing income can shape
At the bottleneck node of a unusual heavyweight, it is further exacerbated by its baneful influence;(2) from functional perspective consider, open source engine its
Distribution is abstract to be substantially " single-instruction multiple-data (SIMD) " mode, and according to the understanding to artificial intelligence application scene, it needs
" multiple-instruction multiple-data (MIMD) " to be more flexible is distributed abstract, and introduce open source engine reduces total system instead
Elasticity and flexibility.
Summary of the invention
It is an object of the invention to: current AI cloud platform is solved using directly stacking open source big data batch processing/stream process
Engine increases system complexity, all very unfavorable for deployment, debugging and maintenance, and introducing open source engine reduces entirety
The problem of elasticity of system and flexibility, propose a kind of artificial intelligence cloud platform.
The technical solution adopted by the invention is as follows:
A kind of artificial intelligence cloud platform, is made of IaaS, PaaS and SaaS three-tier architecture, and IaaS layers are located at bottom, PaaS
Layer is arranged in IaaS layers of upper layer, and SaaS layers are arranged in PaaS layers of upper layer, is successively abstracted using to resource, component and application,
The mode thoroughly decoupled to system constructs this cloud platform, in which:
IaaS layers, using virtualization technology, unified distribution and management bottom hardware resource is used, construct intelligent cloud platform
Infrastructure services;
PaaS layers by carrying out the abstract autonomous Row control engine of building of level Four, and then building to artificial intelligence application scene
Artificial intelligence component and system service, it includes following four parts: data level that it is abstract, which to carry out level Four, to artificial intelligence application scene
Abstract KLoud.cube, functional level are abstracted KLoud.case, service level is abstracted KLoud.chain and optimization level is abstract
KLoud.cycle;
SaaS layer building artificial intelligence application service.
Further, described level Four to be carried out to artificial intelligence application scene abstract detailed process is as follows:
Data level is abstracted KLoud.cube: all business datums being abstracted as to unified data block, it is unified to reach interface;
Functional level is abstracted KLoud.case: all business functions are abstracted as from input cube data to output cube data
Mapping, it is unified that business function is reached into interface;
Service level is abstracted KLoud.chain: all business being abstracted as and are mutually linked group by one or more functions case
At directed acyclic graph;
Optimization level is abstracted KLoud.cycle: closed-loop control is carried out to a business chain, by trial operation to the chain
In involved parameter param carry out tuning.
Further, the directed acyclic graph structures are configured by json script.
Further, business is carried out using the Le Gaoshi mode of building in SaaS layers described to build.
Further, the cloud platform is completed using high concurrent distributed data cache Redis and message-oriented middleware Kafka
System thoroughly decouples, comprising: 1) thoroughly decouples between SaaS layers and PaaS layers and only carry out state transfer to arrange data format;2)
Thoroughly decoupling is carried out between PaaS layers of each component to rely primarily on RESTful micro services and add docker container technique.
Further, full asynchronous system is taken in described PaaS layers of each component design, only carries out state to arrange data format
Transmission.
Further, KLoud.cube and KLoud.case are reconstructed the cloud platform, wherein by KLoud.cube's
Data backup is put into memory pool, and the RESTful micro services framework of KLoud.case is changed to dynamic link library api interface side
Formula provides a monomer elta Force Xtreme.
Further, on the basis of the service level is abstracted KLoud.chain, progress is nested between chain and chain, is
The funcall interface of upper layer application offer different disposal granularity.
In conclusion by adopting the above-described technical solution, the beneficial effects of the present invention are:
1, in the present invention, to IaaS layers, using virtualization technology, unified distribution and management bottom hardware resource are suitable for
The extensive deployment scenarios such as proximal end calculating, private clound, IDC, public cloud.To PaaS layers, four are carried out to artificial intelligence application scene
The abstract autonomous Row control engine of building of grade, and then artificial intelligence component and system service are constructed, service level is abstract
KLoud.chain is abstract for MIMD distribution, with very strong flexibility, is based on the autonomous Row control engine, this is artificial
Intelligent cloud platform may be implemented many flexible and operate effectively, and solving stacking open source engine increases system complexity, for
The all very unfavorable disadvantage of deployment, debugging and maintenance.This cloud platform system has high resiliency, and progress can be facilitated extending transversely, fits
Should be from single machine to the scheme scene of 100,000 medium-scale clusters, cost of reproduction is extremely low.This cloud platform system also has High Availabitity
Property, system service self-healing and online migration, the damage of part machine, which does not influence total system, can be used, and system maintenance is simple, and heat is inserted
Formula damage machine replacement is pulled out, O&M cost is extremely low.Micro services+Service Grid Architecture, can reinforce adapt to proximal end calculating, private clound,
The extensive deployment scenario such as IDC, public cloud, flexible Scheme Choice maximization meet client such as low latency, high security or low
The various crucial requirements such as cost.
2, in the present invention, operation flow chain is conceptualized as directed acyclic graph structures (DAG), and its structure then passes through
The configuration of json script is based on such design, and the engineering technology difficulty of this cloud platform building complicated business logic greatly reduces, because
This can execute online such as multi-modal across media intelligent analysis complex operations under practical scene, improve it and handle complicated industry
The ability for logic of being engaged in.By json script as bridge, cloud platform can more easily develop configuration process modeling platform.
3, in the present invention, on configuration process modeling platform, business is carried out using the Le Gaoshi mode of building in SaaS layers
It builds, user can complete intellectual analysis function required for oneself by the plain mode of similar " taking happy height ".Meanwhile it obtaining
Beneficial to KLoud.chain is abstract to the unification of on-line/off-line business and the data abstraction of KLoud.cube, in process modeling
It is also convenient to that constructed operation flow is debugged and evaluated and tested on platform, it is different that every step implementing result investigation both can be traced
Often, comprehensive assessment also is carried out to performance of MRC process in combination with statistical data.And it is configured by script and carries out Le Gaoshi service logic
It builds, such Le Gaoshi operation flow is abstract to be compressed to the exploitation complete period from requirement investigation to the project implementation one week to one
It is outstanding abstract to ensure that mixed processing in particular for complicated business scenes such as multi-modal across media intelligent analyses within month
Quickly exploitation and deployment are also able to carry out with convergence analysis.Artificial intelligence can be conveniently and efficiently built to SaaS layers in the present invention
Using significantly reducing the technical threshold and development cost of secondary development, take out the software service for having higher universality.
4, in the present invention, KLoud.cube and KLoud.case are reconstructed, wherein the data by KLoud.cube are standby
Part is put into memory pool, and the RESTful micro services framework of KLoud.case is changed to dynamic link library api interface mode, that is, is mentioned
For a monomer elta Force Xtreme, then the operation flow run in cloud platform can be directly migrated to single machine even embedded platform
In, avoid the problem that the system complexity redundancy under proximal end or edge calculations scene.Make this cloud platform that there are other cloud platforms not have
There is the ability for being detached from the operation of IaaS monomer reached.
5, in the present invention, functional level is abstracted KLoud.case and all business functions is abstracted as from input cube data to defeated
The mapping of cube data out can use parameter param in the map, so that it is unified that business function has also been reached to interface.One
A business function can correspond to the docker example of one or more load balancing, benefit from PaaS layers of each component design and take
Full asynchronous system, the function that these docker examples can be isomerization is realized, such as CPU/GPU/FPGA/ASIC version, can
Collection group operatione power is efficiently concentrated output, had by the Heterogeneous Computing for more easily implementing the various platforms such as CPU/GPU/FPGA/ASIC
Effect reduces platform hardware cost.
6, in the present invention, the MIMD distribution for having benefited from chain is abstract, can also carry out between chain and chain embedding
Set further promotes the business processing energy of cloud platform to provide the funcall interface of different disposal granularity for upper layer application
Power.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is cloud platform overall architecture schematic diagram of the present invention;
The intelligent monitoring video process flow diagram of the position Fig. 2 application cloud platform of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention, i.e., described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is logical
The component for the embodiment of the present invention being often described and illustrated herein in the accompanying drawings can be arranged and be designed with a variety of different configurations.
Therefore, the detailed description of the embodiment of the present invention provided in the accompanying drawings is not intended to limit below claimed
The scope of the present invention, but be merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
It should be noted that the relational terms of term " first " and " second " or the like be used merely to an entity or
Operation is distinguished with another entity or operation, and without necessarily requiring or implying between these entities or operation, there are any
This actual relationship or sequence.Moreover, the terms "include", "comprise" or its any other variant be intended to it is non-exclusive
Property include so that include a series of elements process, method, article or equipment not only include those elements, but also
Further include other elements that are not explicitly listed, or further include for this process, method, article or equipment it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including described
There is also other identical elements in the process, method, article or equipment of element.
Part noun used in the present invention is explained below:
RESTful:Representational State Transfer (REST) is a kind of software architecture style, design
Style, rather than standard merely provide one group of design principle and constraint condition, meet the application program of these principles and condition
Or design is exactly RESTful.Most important one REST principle is that the interaction between client and server is between request
Stateless, stateless request can be answered by any available server, this is very suitable for the environment of cloud computing etc.Using journey
Sequence state and function can be divided into various resources, and each resource obtains a unique address using URI, uses standard
HTTP method transmission state between clients and servers.REST simplifies the realization of client and server.
Microservices (micro services): micro services are a kind of framework styles, and a large complicated software application is by one
Or multiple micro services compositions.Each micro services in system can be disposed independently, be loose coupling between each micro services.Each
Micro services, which only focus on, to be completed a task and completes the task well.In all cases, each Charge-de-Mission one
Small professional ability.
Docker container: docker is the application container engine of an open source, allows developer that can be packaged their application
And packet is relied on into a transplantable container, it is then published on the Linux machine of any prevalence, also may be implemented virtual
Change.Container is not have any interface between each other using sandbox mechanism completely.
Kubernetes (k8s): k8s be automate container operation Open Source Platform, these operation include deployment, scheduling and
It is extended between node cluster.Docker can be regarded as to the low level component used inside k8s, it can be with using k8s: (1) automating
The deployment and duplication of container, (2) extend or shrink at any time container scale, and (3) in groups, and provide container organization between container
Load balancing, the new version of (4) easily upgrade applications container, (5) provide container elasticity, replace if container failure
Change it etc..
Service Mesh (service grid environment): the end of the year 2017, the service grid environment technology of non-intrusion type are communicated as between service
Infrastructure layer, from rudiment to moving to maturity.Service grid environment can be likened to application program in other words between micro services
TCP/IP, network call, current limliting, fusing and the monitoring being responsible between service.Service grid environment has the characteristics that following several: (1) answering
With the middle layer communicated between program, (2) lightweight network agent, (3) application program unaware, (4) decouple the weight of application program
Examination/time-out, monitoring, tracking and service discovery.
DevOps:DevOps is the general designation of one group of process, method and system, for promoting exploitation, technology operation and quality
Communication, cooperation and integration between guarantee department.It is a kind of attention " software developer (Dev) " and " IT O&M technology people
The culture, movement or convention of cooperation are linked up between member (Ops) ".Through the process of automation " Software Delivery " and " framework change ",
To enable building, test, publication software more quick, frequent and reliable.
IaaS:Infrastructure as a Service, infrastructure service.Hardware can be outsourced to other
Place is gone, and IaaS can provide over-the-counter server, storage and the network hardware, also can choose rental.It saves maintenance cost and does
Public place can run its application using these hardware at any time.
PaaS:Platform as a Service, platform service.It is sometimes also designated as middleware, all opens
Hair can be carried out in this layer, save time and resource.PaaS can provide the solution of various exploitations and distribution application,
Such as virtual server and operating system etc., the expense on hardware can be saved.
SaaS:Software as a Service, software service.This is the level that ordinary user is in the most contact, in net
Application on network on any one remote server is all to belong to SaaS.
Feature and performance of the invention are described in further detail below.
A kind of artificial intelligence cloud platform, as shown in Figure 1, being made of IaaS, PaaS and SaaS three-tier architecture, IaaS layers are located at
Bottom, PaaS layers are arranged in IaaS layers of upper layer, and SaaS layers are arranged in PaaS layers of upper layer, using to resource, component and application
Successively abstract, the mode thoroughly decoupled to system constructs this cloud platform, in which:
IaaS layers, using virtualization technology, unified distribution and management bottom hardware resource is used, construct intelligent cloud platform
Infrastructure services;
PaaS layers by carrying out the abstract autonomous Row control engine of building of level Four, and then building to artificial intelligence application scene
Artificial intelligence component and system service, it includes following four parts that it is abstract, which to carry out level Four, to artificial intelligence application scene:
Data level is abstracted KLoud.cube: all business datums being abstracted as to unified data block, to reach interface system
One.There is cube data local memory and Redis to cache two backups, and the data between two backups are transmitted and synchronous work
It is automatically performed.
Functional level is abstracted KLoud.case: all business functions are abstracted as from input cube data to output cube data
Mapping, parameter param can be used in the map, so that it is unified that business function has also been reached to interface.
Service level is abstracted KLoud.chain: all business being abstracted as and are mutually linked group by one or more functions case
At directed acyclic graph, directed acyclic graph structures configured by json script (can also using other configuration modes, but
Compare and tend to script configuration mode, such as ini script, xml script, yaml script, protobuf script etc..Json script band
Come advantage, essentially consist in it is preferable with restful Interference fit because http request and response more commonly use json data lattice
Formula, therefore use json script can reduce the work of some Data Format Transforms).The data source header of chain can be one
The headend equipments such as camera or face snap machine, the back-end analysis for real-time online are handled;It is also possible to one from slow
Layer or persistent layer HDFS/ the database even video file of external linkage, picture file or file are deposited, before inquiry etc.
End request response or system debug, evaluation and test etc..In addition, the MIMD distribution for having benefited from chain is abstract, chain and chain
Between can also carry out nesting, to provide the funcall interface of different disposal granularity for upper layer application.MIMD is mainly
Distributed big data processing implementation, at present open source big data frame such as hadoop, spark of mainstream etc. mainly use
SIMD mode, this mode sacrifices the flexibility of data processing in fact, and MIMD mode then very well satisfies flexibility and wants
It asks.
Optimization level is abstracted KLoud.cycle: closed-loop control is carried out to a business chain, by trial operation to the chain
In involved parameter param carry out tuning.
SaaS layers of progress business building construct artificial intelligence application service.
Further, the cloud platform is completed using high concurrent distributed data cache Redis and message-oriented middleware Kafka
System decoupling, there are two levels for this decoupling:
1) it application front end and the decoupling of processing rear end: is thoroughly decoupled between SaaS layers and PaaS layers of system, only to arrange data
Format carries out state transfer.
2) inter-module decouples: thoroughly being decoupled between PaaS layers of system of each components, relies primarily on RESTful in incognito
Business+docker container technique.In addition, full asynchronous system is taken in component design, only with agreement based on the design concept thoroughly decoupled
Data format carries out state transfer.
Further, KLoud.cube and KLoud.case are reconstructed the cloud platform, wherein by KLoud.cube's
Data backup is put into memory pool, and the RESTful micro services framework of KLoud.case is changed to dynamic link library api interface side
Formula provides a monomer elta Force Xtreme.The operation flow then run in cloud platform can be directly migrated to single machine and is even embedded in
In formula platform, the system complexity redundancy under proximal end or edge calculations scene is avoided the problem that.Make this cloud platform that there are other clouds
The ability for being detached from the operation of IaaS monomer that platform does not reach.
Further, business is carried out using the Le Gaoshi mode of building in SaaS layers described to build.In the present invention, pass through json foot
This can more easily develop configuration process modeling platform as bridge, cloud platform, on the platform, use in SaaS layers
The Le Gaoshi mode of building carries out business and builds, and user can be completed by the plain mode of similar " taking happy height " required for oneself
Intellectual analysis function.Meanwhile having benefited from that KLoud.chain is abstract to the unification of online/high line service and KLoud.cube
Data abstraction, be also convenient to that constructed operation flow is debugged and evaluated and tested on process modeling platform, both can be with
The every step implementing result investigation of track is abnormal, also carries out comprehensive assessment to performance of MRC process in combination with statistical data.And matched by script
It sets progress Le Gaoshi service logic to build, such Le Gaoshi operation flow is abstract by the exploitation from requirement investigation to the project implementation
Complete period is compressed within one week to January, outstanding in particular for complicated business scenes such as multi-modal across media intelligent analyses
Abstract ensure that mixed processing and convergence analysis is also able to carry out quickly exploitation and deployment.It can be square to SaaS layers in the present invention
Just artificial intelligence application is quickly built, the technical threshold and development cost of secondary development is significantly reduced, takes out and have
The software service of higher universality.
In the present invention, operation flow chain is conceptualized as directed acyclic graph structures (DAG), and its structure then passes through
The configuration of json script is based on such design, and the engineering technology difficulty of this cloud platform building complicated business logic greatly reduces, because
This can execute online such as multi-modal across media intelligent analysis complex operations under practical scene, improve it and handle complicated industry
The ability for logic of being engaged in.Such as the intelligent monitoring video process flow based on this cloud platform, as shown in Fig. 2, comprehensively utilizing a variety of people
Work intellectual analysis Processing Algorithm achievees the effect that far to win general intelligence monitoring system, such business processing by information integration
Process is difficult efficiently to realize in general AI cloud platform system.Various algorithms in Fig. 2, some of them algorithm solve at present compared with
It is good, but be also presently, there are research bottleneck there are also more algorithms, effect is not good enough, such as person reid, various structures
Change etc., therefore general artificial intelligence platform independently carrys out finishing service with single or certain several algorithm, it is easy to touching reaches this
The performance ceiling of a little algorithms, it is difficult to meet the needs of practical, commercial.In order to break through these bottlenecks, it is necessary to comprehensively utilize
Various intelligent algorithms cooperate, and are complementary to one another, thus reach in effect it is whole promoted, this is namely " multi-modal "
Necessity.And in general cloud platform system, it is very big to complete such work process difficulty.In the present system, due to above
The multiple technologies of use, especially Le Gaoshi are edited, DAG is calculated and schemed abstract, disparate step MIMD distribution, are made it possible to easily
Implement such high-performance multi-modality application.
In the present invention, functional level is abstracted KLoud.case and all business functions is abstracted as from input cube data to output
The mapping of cube data can use parameter param in the map, so that it is unified that business function has also been reached to interface.It is benefited
Full asynchronous system is taken (to be not sole mode, but be preferably mode, be integrated in PaaS layers of each components designs
MIMD can make overall architecture very flexible), a business function can correspond to the docker of one or more load balancing
Example, the function that these docker examples can be isomerization is realized, such as CPU/GPU/FPGA/ASIC version, can be more convenient
The Heterogeneous Computing of the various platforms such as CPU/GPU/FPGA/ASIC is implemented on ground, and collection group operatione power is efficiently concentrated output, is effectively reduced
Platform hardware cost.
In the present invention, to IaaS layers, using virtualization technology, unified distribution and management bottom hardware resource are suitable for close
Hold the extensive deployment scenarios such as calculating, private clound, IDC, public cloud.To PaaS layers, level Four is carried out to artificial intelligence application scene
The abstract autonomous Row control engine of building, and then building artificial intelligence component and system service, service level are abstracted KLoud.chain
It is abstract for MIMD distribution, with very strong flexibility, it is based on the autonomous Row control engine, this artificial intelligent cloud platform can
Operated effectively with realizing many flexible (such as " canary test ", single-step debug, convenient multi-modal artificial intelligence application
Deng), solving stacking open source engine increases system complexity, and the disadvantage all very unfavorable for deployment, debugging and maintenance is (autonomous
Row control engine very lightweight, therefore code maintenance, function renewal etc. are more much more convenient than Open Framework, then due to this
Engine is disparate step+MIMD, therefore can be with single machine or the deployment of independent mirror image, stand-alone debugging, this is just than big data frame of increasing income
It runs directly on cluster, all can be much easier in terms of deployment, debugging and maintenance).This cloud platform system has high resiliency
(because the primary clustering of platform, such as Row control engine KLoud.chain, algoritic module KLoud.case, all using void
Quasi-ization is that containerization mode is disposed and dispatched, and under the support of container orchestration technology, has excellent property extending transversely, cluster
Physical machine configure (quantity) be substantially for component more than virtual level it is noninductive, can be with need to stretch), progress can be facilitated
It is extending transversely, it adapts to from single machine to the scheme scene of 100,000 medium-scale clusters, cost of reproduction is extremely low.This cloud platform system is also
With high availability, (container orchestration technology can monitor each container state, some or certain for system service self-healing and online migration
When a little containers break down, faulty container can be automatically closed and open new same containers be replaced, i.e., service self-healing with
Online migration, to ensure that the high availability of integrity service), the damage of part machine, which does not influence total system, can be used, system
Maintenance is simple, and hot swap type damages machine replacement, and O&M cost is extremely low.Micro services+Service Grid Architecture can be reinforced adapting to proximal end
The extensive deployment scenario such as calculating, private clound, IDC, public cloud, the maximization of flexible Scheme Choice meet client's such as low latency,
The various crucial requirements such as high security or low cost.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (8)
1. a kind of artificial intelligence cloud platform, is made of IaaS, PaaS and SaaS three-tier architecture, IaaS layers are located at bottom, PaaS layers
IaaS layers of upper layer is set, and SaaS layers are arranged in PaaS layers of upper layer, it is characterised in that: using to resource, component and application
Successively abstract, the mode thoroughly decoupled to system constructs cloud platform, in which:
IaaS layers use virtualization technology, unified distribution and management bottom hardware resource, construct the infrastructure of intelligent cloud platform
Service;
PaaS layers are abstracted the autonomous Row control engine of building by carrying out level Four to artificial intelligence application scene, and then construct artificial
Intelligent assembly and system service, it includes following four parts that it is abstract, which to carry out level Four, to artificial intelligence application scene: data level is abstract
KLoud.cube, functional level are abstracted KLoud.case, service level is abstracted KLoud.chain and optimization level is abstracted KLoud.cycle;
SaaS layers for constructing artificial intelligence application service.
2. a kind of artificial intelligence cloud platform according to claim 1, it is characterised in that: described to artificial intelligence application scene
Progress level Four is abstract, and detailed process is as follows:
Data level is abstracted KLoud.cube: all business datums being abstracted as to unified data block, it is unified to reach interface;
Functional level is abstracted KLoud.case: all business functions are abstracted as reflecting from input cube data to output cube data
It penetrates, it is unified that business function is reached interface;
Service level is abstracted KLoud.chain: all business being abstracted as to be mutually linked by one or more functions case forms
Directed acyclic graph;
Optimization level is abstracted KLoud.cycle: closed-loop control is carried out to a business chain, by trial operation to institute in the chain
The parameter param being related to carries out tuning.
3. a kind of artificial intelligence cloud platform according to claim 2, it is characterised in that: the directed acyclic graph structures pass through
Json script is configured.
4. a kind of artificial intelligence cloud platform according to claim 2, it is characterised in that: use Le Gaoshi in SaaS layers described
The mode of building carries out business and builds.
5. a kind of artificial intelligence cloud platform according to claim 1, it is characterised in that: the cloud platform is using high concurrent point
Cloth data buffer storage Redis and message-oriented middleware Kafka thoroughly decouple to complete system, comprising: 1) SaaS layers and PaaS layer
Between thoroughly decouple only with arrange data format carry out state transfer;2) between PaaS layers of each component carry out thoroughly decoupling mainly according to
RESTful micro services are relied to add docker container technique.
6. a kind of artificial intelligence cloud platform according to claim 1 or 4, it is characterised in that: described PaaS layers of each component
Full asynchronous system is taken in design, only carries out state transfer to arrange data format.
7. a kind of artificial intelligence cloud platform according to claim 1 or 2, it is characterised in that: the cloud platform will
KLoud.cube and KLoud.case are reconstructed, wherein the data backup of KLoud.cube is put into memory pool, it will
The RESTful micro services framework of KLoud.case is changed to dynamic link library api interface mode, provides a monomer elta Force Xtreme.
8. a kind of artificial intelligence cloud platform according to claim 1 or 2, it is characterised in that: abstract in the service level
On the basis of KLoud.chain, progress is nested between chain and chain, provides the function tune of different disposal granularity for upper layer application
Use interface.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910112462.5A CN109828831B (en) | 2019-02-12 | 2019-02-12 | Artificial intelligence cloud platform |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910112462.5A CN109828831B (en) | 2019-02-12 | 2019-02-12 | Artificial intelligence cloud platform |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109828831A true CN109828831A (en) | 2019-05-31 |
CN109828831B CN109828831B (en) | 2020-10-16 |
Family
ID=66863650
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910112462.5A Active CN109828831B (en) | 2019-02-12 | 2019-02-12 | Artificial intelligence cloud platform |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109828831B (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110427173A (en) * | 2019-07-26 | 2019-11-08 | 大唐融合通信股份有限公司 | A kind of artificial intelligence synthesis business platform |
CN110706392A (en) * | 2019-10-14 | 2020-01-17 | 重庆紫光华山智安科技有限公司 | Security platform test method and device, storage medium and server |
CN110909610A (en) * | 2019-10-26 | 2020-03-24 | 湖北讯獒信息工程有限公司 | Accurate age identification method based on artificial intelligence |
CN110930117A (en) * | 2019-11-21 | 2020-03-27 | 中国银行股份有限公司 | Artificial intelligence micro service system |
CN110990121A (en) * | 2019-11-28 | 2020-04-10 | 中国—东盟信息港股份有限公司 | Kubernetes scheduling strategy based on application portrait |
CN111400071A (en) * | 2020-04-10 | 2020-07-10 | 深圳新致软件有限公司 | Artificial intelligence service docking method, system and equipment |
CN111401566A (en) * | 2020-03-19 | 2020-07-10 | 中国建设银行股份有限公司 | Machine learning training method and system |
CN111427687A (en) * | 2020-03-23 | 2020-07-17 | 深圳市中盛瑞达科技有限公司 | Artificial intelligence cloud platform |
CN111477335A (en) * | 2020-04-01 | 2020-07-31 | 江苏省测绘工程院 | Epidemic situation space-time big data platform based on micro-service architecture and construction method |
CN111897556A (en) * | 2020-06-29 | 2020-11-06 | 江苏柏勋科技发展有限公司 | Video monitoring service system based on cloud computing |
CN112201099A (en) * | 2019-07-08 | 2021-01-08 | 苏州易学在线文化传播有限公司 | Online education platform |
CN112667360A (en) * | 2020-12-31 | 2021-04-16 | 宝付网络科技(上海)有限公司 | Cloud platform system based on Kubernetes and docker unified scheduling |
CN113254525A (en) * | 2021-02-07 | 2021-08-13 | 清华大学 | Collaborative manufacturing engine system constructed based on industrial PAAS and implementation method |
CN115167292A (en) * | 2021-04-12 | 2022-10-11 | 清华大学 | Intelligent factory operating system based on industrial Internet architecture |
CN117216148A (en) * | 2023-11-09 | 2023-12-12 | 三峡科技有限责任公司 | Alkaline water electrolysis hydrogen production system operation maintenance platform system |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100275077A1 (en) * | 2009-04-22 | 2010-10-28 | Mentor Graphics Corporation | At-Speed Scan Testing With Controlled Switching Activity |
CN103312727A (en) * | 2012-03-06 | 2013-09-18 | 创业软件股份有限公司 | Method for cloud computing deployment in the field of cooperative services of regional medical care |
US20150188927A1 (en) * | 2013-03-15 | 2015-07-02 | Gravitant, Inc | Cross provider security management functionality within a cloud service brokerage platform |
CN108270855A (en) * | 2018-01-15 | 2018-07-10 | 司中明 | A kind of method of platform of internet of things access device |
CN108306981A (en) * | 2018-03-08 | 2018-07-20 | 江苏省广播电视总台 | The dual isomery PaaS redundancy approach of media cloud platform service layer safety can be enhanced |
CN108667850A (en) * | 2018-05-21 | 2018-10-16 | 济南浪潮高新科技投资发展有限公司 | A kind of artificial intelligence service system and its method for realizing artificial intelligence service |
CN109254759A (en) * | 2018-08-31 | 2019-01-22 | 重庆戴昂科技有限公司 | Low code hardware and software platform operation flow configures system |
-
2019
- 2019-02-12 CN CN201910112462.5A patent/CN109828831B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100275077A1 (en) * | 2009-04-22 | 2010-10-28 | Mentor Graphics Corporation | At-Speed Scan Testing With Controlled Switching Activity |
CN103312727A (en) * | 2012-03-06 | 2013-09-18 | 创业软件股份有限公司 | Method for cloud computing deployment in the field of cooperative services of regional medical care |
US20150188927A1 (en) * | 2013-03-15 | 2015-07-02 | Gravitant, Inc | Cross provider security management functionality within a cloud service brokerage platform |
CN108270855A (en) * | 2018-01-15 | 2018-07-10 | 司中明 | A kind of method of platform of internet of things access device |
CN108306981A (en) * | 2018-03-08 | 2018-07-20 | 江苏省广播电视总台 | The dual isomery PaaS redundancy approach of media cloud platform service layer safety can be enhanced |
CN108667850A (en) * | 2018-05-21 | 2018-10-16 | 济南浪潮高新科技投资发展有限公司 | A kind of artificial intelligence service system and its method for realizing artificial intelligence service |
CN109254759A (en) * | 2018-08-31 | 2019-01-22 | 重庆戴昂科技有限公司 | Low code hardware and software platform operation flow configures system |
Non-Patent Citations (3)
Title |
---|
DAVID S. LINTHICUM ET AL: "Making Sense of AI in Public Clouds", 《IEEE CLOUD COMPUTING 》 * |
YAWEI ZHAO ET AL: "Research on parallel query technology of pXCube model based on cloud pattern", 《 2011 IEEE INTERNATIONAL CONFERENCE ON CLOUD COMPUTING AND INTELLIGENCE SYSTEMS》 * |
沈玉姗: "区块链落地", 《二十一世纪商业评论》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112201099A (en) * | 2019-07-08 | 2021-01-08 | 苏州易学在线文化传播有限公司 | Online education platform |
CN110427173B (en) * | 2019-07-26 | 2023-02-03 | 大唐融合通信股份有限公司 | Artificial intelligence integrated service platform |
CN110427173A (en) * | 2019-07-26 | 2019-11-08 | 大唐融合通信股份有限公司 | A kind of artificial intelligence synthesis business platform |
CN110706392B (en) * | 2019-10-14 | 2021-10-26 | 重庆紫光华山智安科技有限公司 | Security platform test method and device, storage medium and server |
CN110706392A (en) * | 2019-10-14 | 2020-01-17 | 重庆紫光华山智安科技有限公司 | Security platform test method and device, storage medium and server |
CN110909610A (en) * | 2019-10-26 | 2020-03-24 | 湖北讯獒信息工程有限公司 | Accurate age identification method based on artificial intelligence |
CN110930117A (en) * | 2019-11-21 | 2020-03-27 | 中国银行股份有限公司 | Artificial intelligence micro service system |
CN110990121A (en) * | 2019-11-28 | 2020-04-10 | 中国—东盟信息港股份有限公司 | Kubernetes scheduling strategy based on application portrait |
CN111401566A (en) * | 2020-03-19 | 2020-07-10 | 中国建设银行股份有限公司 | Machine learning training method and system |
CN111401566B (en) * | 2020-03-19 | 2024-05-03 | 中国建设银行股份有限公司 | Machine learning training method and system |
CN111427687A (en) * | 2020-03-23 | 2020-07-17 | 深圳市中盛瑞达科技有限公司 | Artificial intelligence cloud platform |
CN111477335A (en) * | 2020-04-01 | 2020-07-31 | 江苏省测绘工程院 | Epidemic situation space-time big data platform based on micro-service architecture and construction method |
CN111400071A (en) * | 2020-04-10 | 2020-07-10 | 深圳新致软件有限公司 | Artificial intelligence service docking method, system and equipment |
CN111400071B (en) * | 2020-04-10 | 2024-01-09 | 深圳新致软件有限公司 | Artificial intelligence service docking method, system and equipment |
CN111897556A (en) * | 2020-06-29 | 2020-11-06 | 江苏柏勋科技发展有限公司 | Video monitoring service system based on cloud computing |
CN112667360A (en) * | 2020-12-31 | 2021-04-16 | 宝付网络科技(上海)有限公司 | Cloud platform system based on Kubernetes and docker unified scheduling |
CN113254525A (en) * | 2021-02-07 | 2021-08-13 | 清华大学 | Collaborative manufacturing engine system constructed based on industrial PAAS and implementation method |
CN115167292A (en) * | 2021-04-12 | 2022-10-11 | 清华大学 | Intelligent factory operating system based on industrial Internet architecture |
CN115167292B (en) * | 2021-04-12 | 2024-04-09 | 清华大学 | Intelligent factory operating system based on industrial Internet architecture |
CN117216148A (en) * | 2023-11-09 | 2023-12-12 | 三峡科技有限责任公司 | Alkaline water electrolysis hydrogen production system operation maintenance platform system |
CN117216148B (en) * | 2023-11-09 | 2024-02-06 | 三峡科技有限责任公司 | Alkaline water electrolysis hydrogen production system operation maintenance platform system |
Also Published As
Publication number | Publication date |
---|---|
CN109828831B (en) | 2020-10-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109828831A (en) | A kind of artificial intelligence cloud platform | |
de Assuncao et al. | Distributed data stream processing and edge computing: A survey on resource elasticity and future directions | |
Madsen et al. | Reliability in the utility computing era: Towards reliable fog computing | |
Kakivaya et al. | Service fabric: a distributed platform for building microservices in the cloud | |
US11544149B2 (en) | System and method for improved fault tolerance in a network cloud environment | |
Mikkilineni et al. | The Turing O-Machine and the DIME Network Architecture: Injecting the Architectural Resiliency into Distributed Computing. | |
De Benedetti et al. | JarvSis: a distributed scheduler for IoT applications | |
US11050624B2 (en) | Method and subsystem that collects, stores, and monitors population metric data within a computer system | |
Bedini et al. | Modeling performance of a parallel streaming engine: bridging theory and costs | |
Lu et al. | Assessing MapReduce for internet computing: a comparison of Hadoop and BitDew-MapReduce | |
Ramon-Cortes et al. | A survey on the Distributed Computing stack | |
US11184244B2 (en) | Method and system that determines application topology using network metrics | |
Deng et al. | Cloud-native computing: A survey from the perspective of services | |
Branson et al. | CLASP: C ol laborating, A utonomous S tream P rocessing Systems | |
Wei et al. | OpenCluster: a flexible distributed computing framework for astronomical data processing | |
Chow et al. | {DQBarge}: Improving {Data-Quality} Tradeoffs in {Large-Scale} Internet Services | |
Sugiki et al. | An extensible cloud platform inspired by operating systems | |
Leite | A user-centered and autonomic multi-cloud architecture for high performance computing applications | |
Douglis et al. | Multi-site cooperative data stream analysis | |
Abawajy | Autonomic job scheduling policy for grid computing | |
US20210027091A1 (en) | Multi-domain monitoring services for intelligent infrastructure automation | |
Higashino | Complex event processing as a service in multi-cloud environments | |
Nivethitha et al. | Survey on architectural design principles for edge oriented computing systems | |
MEMON et al. | A technique to differentiate clustered operating systems | |
Usai | Delivering Resilient Virtualized Services in Smart Grid Environments |
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 |