CN108228347A - The Docker self-adapting dispatching systems that a kind of task perceives - Google Patents
The Docker self-adapting dispatching systems that a kind of task perceives Download PDFInfo
- Publication number
- CN108228347A CN108228347A CN201711395307.6A CN201711395307A CN108228347A CN 108228347 A CN108228347 A CN 108228347A CN 201711395307 A CN201711395307 A CN 201711395307A CN 108228347 A CN108228347 A CN 108228347A
- Authority
- CN
- China
- Prior art keywords
- container
- service
- controller
- resource
- scheduler
- 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.)
- Pending
Links
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
-
- 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
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The Docker self-adapting dispatching systems that a kind of task perceives, Docker self-adapting dispatching method flows are：Data collector acquires the load of each container and CPU, memory system resource utilization；Performance modeling device is based on queueing theory, using resource utilization as benchmark, builds application performance model, portrays the incidence relation of load and response time；Response time fallout predictor estimates that estimation is to be expected to the condition of laying in a coffin to meet prediction and actual measurement response error when being run by Kalman filter to performance model parameter；Feedforward controller analyzes residual error mean value and variance by fuzzy controller, obtains the feedforward adjusted value of Kalman filter control parameter；Container scheduler judges whether the predicted value of response time has violated application service quality, and is scheduled according to dispatching algorithm；Container scheduler performs container expansion, contraction or migration.
Description
Technical field
The invention belongs to field of cloud computer technology, and the Docker that a kind of task more particularly under micro services framework perceives is certainly
Adaption scheduling system.
Background technology
Micro services framework embodies the design philosophy of the Internet, applications, and core concept is that fine granularity module is divided, serviced
Change interface encapsulation, lightweight communication interaction, micro services have good autgmentability, are increasingly becoming construction the Internet, applications in itself
Mainstream architecture mode, but from the viewpoint of operation and maintenance, cope with typical internet mutation load scene, ensure application
Service quality still faces the challenge.
In recent years, Lightweight Container technology is come into being, logical abstraction of the container as physical resource, has resource occupation
Less, the features such as resource provision is fast is suitble to the Internet, applications pattern of workload mutation, particularly towards the new of micro services framework
Type services pattern.However, existing work is limited to physical machine and virtualized environment or resource is difficult to elastic supply or resource provision
Timeliness is poor, it is difficult to cope with load sudden change scene.
In existing cloud framework, scheduling of resource universal process is generally divided into four steps：Resource request, resource detection,
Resource selection, monitoring resource.To be managed collectively these virtual resources, each available virtual resource is abstracted into a schedulable
Unit (Unit), each Unit are determined that wherein C represents CPU sizes by a two tuple V (C, M), and M represents memory size.Then one
The virtual Unit numbers of platform physical server Si most multipotencys are：Forms data center available resources sum isIntroduce the thought of grouping simultaneously, will be grouped with the resource of identical handling characteristics, so as to avoid because
United Dispatching largely consumption of the random resource to cloud platform performance, as schedulable unit introduce new attribute F, each Unit by
One triple V (C, M, F) determines that wherein F represents the central characteristics of resource, with reference to dividing above production task type terminal
Class includes the virtual resource of three classes different characteristics under production task type scene altogether.When user logins cloud platform, and the input phase should be able to
Power demand completes resource request.Cloud platform is connected to after user resources request, can be first by being mounted on every server
Data collection module probe node load information on node is (i.e.：Calculate the available of the remaining different characteristics of each resource node
Resource quantity) and server response time.
Then using dynamic priority (Priority) algorithm, Priority value highest nodes are selected, will thereon with user
The virtual resource that characteristic matches consigns to user.Server is grouped by dynamic priority (Priority) algorithm, to each
Group defines different priority, and user's request can distribute to the highest server group of priority (in same group, using poll or ratio
Rate algorithm, the request of distributing user).The Priority values of every group of server are responded by the loading condition of present node with server
Time codetermines.The core concept of dynamic priority algorithm is user's request by present load minimum and the shortest resource of response
Node carries.Priority=Unit total × F+LT × μ, wherein F represent user demand characteristic weight, and LT represents server
Response time, μ represent time priority power conversion ratio.When user asks to reach, Priority values dynamic updates, exactly this machine
System so that dynamic priority (Priority) algorithm is difficult to cope with the situation of load sudden change.
Existing patent document includes, and is disclosed application No. is CN201710461892.9 files " a kind of pre- based on load
The Docker Swarm cluster resource method for optimizing scheduling of survey, belongs to computer system virtualization technical field.The present invention utilizes
The resource history usage amount of the api interface function of Docker daemon, periodically collection vessel；Take ARIMA-RBF models
To carry out modeling and forecasting to the resource history usage amount of container, obtain resource future usage amount, gather the current service condition of resource
Resource SC service ceiling and resources use right limit are adjusted；And determine container to money according to the resource history usage amount of container
Degree is biased in the use in source；And when cluster starts a new container according to the container and the set of node for meeting container resource requirement
On resource using deviation degree, selecting one to add in resource after the container, to dispose this using most balanced node is biased to new
Container；Technical solution provided by the invention improves the resource utilization of Docker Swarm clusters, and can promote each container
Actual motion performance ".
Patent document application No. is CN201710289847.X disclose " a kind of Container Management platform, including：Scheduling
Device, for the application schedules frame write based on Mesos Restful API, the life cycle management applied for container；Monitoring
Alarm system, for the performance indicator of monitoring of containers and the health status of application；Log processing system, for log searching and day
Will counts；Delivery system is used to implement the publication and rollback of application.A kind of Container Management platform provided by the invention, Ke Yifei
The often quickly position of one service of positioning, and can also flexibly being set efficiently using the multi-tenant management under cluster resource
Monitoring and alarming system is put, and the daily record provided under associated context is checked ".
Application No. is the patent documents of CN201710221393.2, disclose a kind of " host prison based on micro services framework
Control system, wherein, including：Jenkins continues deployed environment module, for receiving the upgrade code of each micro services, and it is corresponding
Each micro services are carried out with code update, test and deployment and new micro services are installed in server-side；Micro services module, storage
There are multiple micro services, each micro services are used to implement a standalone feature of server-side, and function services are provided to plugin manager,
And the operate interface of server-side is provided system manager；Plugin manager is used to provide basic environment for the operation of each plug-in unit
And it is managed；Plug-in unit is the final execution unit of terminal security monitoring function, for the security strategy that reading service end issues,
And management and control is carried out to terminal according to described in security strategy ".
Application No. is the patent documents of CN201710217944.8, disclose a kind of " application example elasticity based on container
Flexible implementation method, apparatus and system, including：Determine destination host node corresponding with intended application and target container；Really
The index of fixed elastic telescopic strategy corresponding with intended application；From destination host node, finger corresponding with target container is obtained
Mark actual value；If according to index actual value and elastic telescopic strategy, judgement needs to adjust the number of containers of institute's intended application
It is whole, then to Docker Swarm components send number of containers adjust instruction so that Docker Swarm components to number of containers into
Row adjustment；As it can be seen that in the present embodiment, matched by the index actual value of acquisition with scheduled elastic telescopic strategy, it can
Automatic trigger Swarm cluster components are scheduled establishment or delete container, realize automatic expansion of the application example under the conditions of a certain
It is perhaps flexible, promote the high reliability applied based on container ".
However, above-mentioned technical solution can not all solve the problems, such as burst load proposed by the present invention.It is of the present invention
Docker is an application container engine increased income, allow developer can be packaged they application and rely on packet it is removable to one
It in the container of plant, is then published on the Linux machines of any prevalence, can also realize virtualization, container is complete using husky
Punch-out equipment system does not have any interface between each other.
Invention content
Present invention introduces carrier of the container as the Internet, applications, the characteristics of using grade resource provision in its second, one kind is provided
The adaptive container resource provision system that the Docker self-adapting dispatching systems and a kind of task that task perceives perceive, meets
The demand of load sudden change.
The Docker self-adapting dispatching systems that a kind of task perceives, the system include data collector, performance modeling device, sound
Versus time estimator, feedforward controller, container scheduler and several containers are answered,
Docker self-adapting dispatching method flows are：
1. data collector acquires the load of each container and CPU, memory system resource utilization；
2. performance modeling device be based on queueing theory, using step 1. middle resource utilization be used as benchmark, structure application performance model,
Portray the incidence relation of load and response time；
3. response time fallout predictor is estimated when being run by Kalman filter to performance model parameter, estimation be with
Meet prediction and actual measurement response error is expected to the condition of laying in a coffin；
4. feedforward controller analyzes residual error mean value and variance by fuzzy controller, Kalman filter control parameter is obtained
Feedforward adjusted value；
5. container scheduler judges whether the predicted value of response time has violated application service quality, and according to dispatching algorithm
It is scheduled；
After 6. container scheduler performs container expansion, contraction or migration, step is continued to execute 1., forming method closed loop.
The adaptive container resource provision system that a kind of task perceives, including automatic deployment module, resource flexible scheduling mould
Block, service register and discovery module,
Automatic deployment module is made of template warehouse and application deployment device, and the deployment mould that can refer to is provided in template warehouse
Plate and application service, user select concrete application and its version number according to demand for services, when template warehouse cannot meet user's clothes
During business demand, user according to the autonomous drawing template establishment of template style,
Resource flexible scheduling module is made of the scheduler of main controlled node and the controller of child node, the scheduler according to
The scheduling request that the controller of user and child node transmits carries out overall scheduling distribution；
Service registration and discovery module realize that inside is tree-shaped using level using distributed consistency key assignments storage system
The process of storage organization, service registration and discovery is：When child node has new service creation, controller can be according to the service
The network address of service and port numbers are registered to the service information store of main controlled node, memory analysis clothes by configuration information
The configuration information of business classifies with being combined automatically to service,
The adaptive container resource provision working-flow that task perceives is as follows：
System uses host-guest architecture, and user is selected or created in template warehouse using required configuration, then assembled one
It puts template to be transmitted to using deployment device, configuration template be carried out application configuration information using deployment device after fault-tolerant verification, combination
Global scheduler is transmitted to, global scheduler selects suitable child node to carry out according to the resource service condition of current each child node
Service arrangement, and by information on services persistent storage after the controller of child node receives traffic order, is specifically dispatched,
On the other hand, the information of controller also real-time collecting service is reported to the scheduler of main controlled node, when there is new demand servicing
During generation, service registration can be carried out from the memory of trend main controlled node.
The present invention is a kind of adaptive feed systems of the Docker towards micro services framework, is restrained based on Kalman filtering
Soon, the characteristics of need not saving historical data, service response time is predicted, and the bullet of the realization resource according to prediction result
Property supply, make up the deficiency that existing research work is difficult to cope with burst load.Wherein, task sensor model is assessment application resource
Demand, the foundation for realizing elastic supply.Task perception is typically required for traffic monitoring and early warning system, ensures in the big stream of reply
Realize that Qos is ensured when amount, the user access request of high concurrent.
The present invention establishes the Docker self-adapting dispatching systems that a kind of task perceives, using a kind of service quality
(Quality of Service, QoS) is sensitive, container resource-adaptive dispatching method based on feedforward, the reality that task perceives
It is now main that workload, resource utilization and the incidence relation of response time are portrayed using queueing theory, so as to build application performance
Model.Wherein, the response time is predicted (feedforward controller) using Fuzzy Adaptive Kalman Filtering, and prediction result is violated
QoS is the foundation for triggering the scheduling of container resource-adaptive.
Micro services container resource-adaptive Scheduling Framework proposed by the present invention, using container lightweight feature, improves money
The actual effect of source supply.The characteristics of using Fuzzy Adaptive Kalman Filtering Fast Convergent, the prediction for improving mutation load are accurate
True property and the validity of resource provision.
Description of the drawings
Detailed description below, above-mentioned and other mesh of exemplary embodiment of the invention are read by reference to attached drawing
, feature and advantage will become prone to understand.In the accompanying drawings, if showing the present invention's by way of example rather than limitation
Dry embodiment, wherein：
Fig. 1 self-adapting dispatching system structure charts of the present invention.
Fig. 2 adaptive container resource provision system construction drawings of the present invention.
Specific embodiment
The present invention task perceive Docker self-adapting dispatching methods overall framework and flow as shown in attached drawing Fig. 1,
This method builds task sensor model with parameters such as the resource utilization of each micro services and loads, is filtered using adaptive Kalman
Wave device predicts service response time, and passes through fuzzy logic and prediction model is adjusted in real time, finally with Service Quality
Whether amount breaks a contract as container Scheduling criteria, achievees the purpose that resource elastic supply.
The Docker self-adapting dispatching method flows that the task of the present invention perceives can be described as：
1. data collector acquires the system resources utilization rates such as load and CPU, the memory of each container；
2. performance modeling device be based on queueing theory, using step 1. middle resource utilization be used as benchmark, structure application performance model,
Portray the incidence relation of load and response time；
3. response time fallout predictor is estimated when being run by Kalman filter to performance model parameter, estimation be with
Meet prediction and actual measurement response error is expected to the condition of laying in a coffin；
4. feedforward controller analyzes residual error mean value and variance by fuzzy controller, Kalman filter control parameter is obtained
Feedforward adjusted value；
5. container scheduler judges whether the predicted value of response time has violated application service quality, and according to dispatching algorithm
It is scheduled；
6. after performing container expansion, contraction or migration, step is continued to execute 1., forming method closed loop.
(1) the application performance model being lined up based on Jackson networks
Jackson open loop networks are to be suitble to the application performance model of micro services framework, and user's request can redirect in node,
By the processing of related micro services, user is finally responded to.When some micro services is there are when multiple examples, using Round Robin
(round-robin scheduling) strategy.Due to different (such as intensive collection of CPU intensive collection type, I/O of the resource preference of micro services
Type), cause the preference resource that container occurs different, define preference resource as utilization rate highest in container CPU, memory, magnetic disc i/o
Resource.
uj∈ [0,1) be micro services j preference resource utilization；u0jRefer to when micro services j is inclined under immunization with gD DNA vaccine
Good resource utilization；γjiRefer to the number of concurrent of i-th of container of micro services j, i.e., the number of request of arrival per second meets Poisson and arrives
Up to process；TjiRefer to the service processing time of i-th of container of micro services j；TjWhen referring to the average service processing of micro services j
Between；D refers to that user asks the overall network transmission time of stream f；B refers to the response time of service flow f；τjBe service j load with
The related coefficient of resource utilization.Had according to Jackson network flows equation and network performance equation：
Wherein, uj, u0j, γji, B is obtained by monitoring, τjIt is the empirical value provided according to historical data, Tji, d is
It is difficult to what is monitored, needs to be estimated by prediction.So-called elastic supply refers to response time B in relatively-stationary interval
Under the premise of, the resource requirement of application.As it can be seen that Tji, d is the key element for carrying out adaptive resource supply.
(2) prediction of the adaptive Kalman filter algorithm to the response time
Kalman filtering algorithm is a kind of optimum linearity method for estimating state proposed by Kalman in nineteen sixty, Chang Beiyong
Field is predicted in track following.Its advantage is to solve linear filtering problem using recursive method, only needs current measured value
State estimation can be just carried out with the estimated value in previous sampling period.
Adaptive Kalman filter (adaptive Kalman filtering, AFK) original equation is as follows：
X (k+1)=F (k) X (k)+Q (k), (3)
Z (k)=H (k) X (k)+R (k), (4)
Wherein, X (k) is prediction matrix, represents the matrix of service processing time and overall delay；Z (k) is the state square of X (k)
Battle array；H (k) is responsible for the multi-C vector of observables being transformed into the multi-C vector of value to be predicted, is the resource profit by application example
With the matrix formed of rate, load and response time；Q (k) is procedure activation noise covariance matrix, and R (k) is measurement noise
Covariance matrix, it is considered that this 2 noise matrixes should be set as zero-mean white noise.But load variation is not often true
It is fixed, such as load sudden change scene, so in order to make the supply of the flexible resource of system that there is real-time, procedure activation noise covariance
Matrix and measurement noise covariance matrix should be adjusted adaptively at any time.Judge whether wave filter needs newer foundation just
It is monitoring residual error, ideally residual error is zero-mean white noise, i.e., wave filter can be perfect adaptive, if residual error is not zero
Mean value white noise then illustrates that error occurs in filter prediction.Since residual variance, residual error mean value are related to Q and R, can pass through
Estimate residual variance and mean value, then carry out fuzzy reasoning, finally adjust the value of U and T, having reached fits Kalman filtering algorithm
Answer the purpose of Time variable structure.
The present invention establishes residual variance and mean value Triangleshape grade of membership function and fuzzy rule using TS fuzzy logic systems
Then, the adaptive Kalman filter algorithm based on fuzzy logic be by the way of feedforward control, it is real-time according to filter forecasting value
Filtering Model parameter is adjusted.The present invention adjusts noise and procedure activation matrix by feedback control in real time, to reach certainly
Adapt to correct the purpose of filtering parameter.
(3) container scheduling strategy
The present invention, can be right according to load and resource service condition by the analysis to response time prediction model method
Container carries out Real-Time Scheduling, to ensure the smoothly output response time.The container scheduling strategy of the present invention mainly has 3 kinds：
1) container migrates.The main reason for it is generated is container own resources using not reaching on resource constraint
It limits rather than because the total resources of host will reach the limitation upper limit, does not need to be extended container at this time.Control
Device only needs service name being reported to scheduler, scheduler will according to resource type and cluster resource service condition to container into
Row migration.Container transition process is approximately as container persistence is mirrored by controller first, this step will preserve current answer
With state, then scheduler sends out traffic order, and the controller for receiving traffic order will be generated according to traffic order from mirror image
1 new container, then feeds back to service register module, to complete service discovery.Original container will be deleted by controller,
Service register module is equally also reported to, completes service deregistration.
2) container expansion.The main reason for it is generated is that a certain resource utilization of container itself uses the limitation for having arrived limiter
Value, carrying out container migration at this time can not solve the problems, such as, to ensure the average response time of the container, it is necessary to which the container is carried out
Extension.The process of container expansion is approximately as scheduler selects to close according to the report information of controller in current cluster first
Suitable node, which is sent out, creates container order, and the configuration information of container is transferred to the controller in node, and controller creates new
The network address according to container and port numbers are subjected to service registration automatically after container.
3) containers shrink.The main reason for it is generated is that the resource service condition of each example of application is below estimated
Value, needs to cut the example quantity of the application at this time.The process of containers shrink is approximately as scheduler is according to controller first
Report information, judges whether the application is shunk, and will determine that result feeds back to controller, and controller receives scheduling life
Automatically service deregistration is carried out after order.
The adaptive container resource provision system and device that task perceives under micro services framework is divided into 3 parts：Automatic deployment
Module, resource flexible scheduling module, service register and discovery module.Its system construction drawing is as shown in attached drawing Fig. 2.
The main working process for the adaptive container resource provision system and device that task perceives is as follows：System is tied using principal and subordinate
Structure, user are selected or are created in template warehouse using required configuration, then one group of configuration template is transmitted to using deployment device, should
Application configuration information is transmitted to global scheduler, overall scheduling after fault-tolerant verification, combination are carried out to configuration template with deployment device
Device selects suitable child node to carry out service arrangement, and information on services is held according to the resource service condition of current each child node
Longization stores, and after the controller of child node receives traffic order, is specifically dispatched, and another aspect, controller can also be received in real time
The information of collection service is reported to the scheduler of main controlled node, can be from the memory of trend main controlled node when there is new demand servicing generation
Carry out service registration.
(1) automatic deployment module
Automatic deployment module is made of template warehouse and application deployment device, and the portion that largely can refer to is provided in template warehouse
Template and application service are affixed one's name to, user can select concrete application and its version number according to demand for services, when template warehouse cannot expire
During sufficient users service needs, user can be according to the autonomous drawing template establishment of template style.
Using a series of model informations that deployment device is transmitted according to template warehouse, the association for servicing required inter-module is closed
System is analyzed, and analysis result is persisted in configuration file, these information will determine the deployment of each service and open
The basic dependence of dynamic sequence and service register and discovery.On the other hand, system can analyze the configuration file of Self -adaptive, into
While row fault-tolerance is verified, corresponding Jackson network queuing models can be also generated, which is that system carries out flexible resource
The key point of scheduling.
(2) resource flexible scheduling module
Resource flexible scheduling module is made of the scheduler of main controlled node and the controller of child node.Scheduler can according to
The scheduling request that family and controller transmit carries out overall scheduling distribution, and the main dispatching principle of scheduler has：Ensure service operation
Priority, by the service of CPU intensive type and I/O intensities service interspersion, ensure that the fair use resource of each service etc. is regular.
Since container is different from conventional virtual machine, resource constraint can not be done to it when creating container, allow its basis
Resource is voluntarily applied for or discharged to the operation demand of service, can maximumlly utilize the physical resource of host in this way, but adopt
The problem of taking this mode that can introduce resource contention simultaneously.Such as when the memory usage amount of multiple containers in host jumps simultaneously
When, it may result in operating system and force to stop container (out of memory kill)；When the network transmission volume of some container
When very big, may result in main controlled node can not receive the heartbeat packet of child node, and system call is led to problems such as to stagnate.Institute
To need to add 1 resource constraint module in the controller, ensure the basic resources of system process.In addition, when each in host
The resource usage amount of a service is all very high, and when total usage amount has reached theoretical limits value, is that service is migrated, still
Service is extendedServices migrating has certain delay and network overhead, although service extension expense is smaller, is difficult
Ensure that original service can continue to keep the state of high resource utilization, it is therefore possible to after service extension, and taken
Business is shunk, and system is caused frequently to be scheduled.To avoid this situation, the load of the controller meeting each service of real-time collecting, resource
The output response time of system is predicted in real time by Kalman filtering using data such as, performances, is then patrolled by fuzzy
Volume theory is classified and reasoning, and whether the service of finally obtaining needs to carry out services migrating or to carry out service flexible.
(3) service register and discovery module
Service registration and discovery module realize that inside is tree-shaped using level using distributed consistency key assignments storage system
Storage organization, can the very efficiently incidence relation between storage service.The process of service registration and discovery can be described as, group
When node has new service creation, controller can be noted the network address of service and port numbers according to the configuration information of the service
Volume gives the service information store of main controlled node, and the configuration information of memory meeting Analysis Service is classified automatically to service and group
It closes.
Load is to influence the principal element of application resource demand, and the present invention is built negative by Jackson network queuing models
The incidence relation with QoS is carried, and whether is broken a contract as the foundation of resource provision using QoS.Specifically, the present invention first with
The load of Jackson network queuing theories structure, resource utilization and the performance model of response time, then using adaptive Kalman
Filtering algorithm predicts the unknown parameter in model, and passes through the control in fuzzy logic (feedforward controller) correcting filter
Parameter processed improves response time forecasting accuracy to reach, ensures the purpose of QoS.
With the development of micro services technology, significant change, cloud service resource section also has occurred in cloud service Floor layer Technology framework
Point quantity is more and more.Meanwhile number of users is more, industry is popularized, demand for services is more, and timeliness it is expected high, data magnanimity and more
The features such as sample, is increasingly apparent.Existing task scheduling algorithm is difficult to meet the needs of task scheduling under the new situation.This method is with every
The parameters such as the resource utilization of a micro services and load build task sensor model, using adaptive Kalman filter to service
Whether the response time is predicted, and passes through fuzzy logic and prediction model is adjusted in real time, finally broken a contract with service quality
As container Scheduling criteria, achieve the purpose that resource elastic supply.The technical solution provided according to the present invention improves the response time
Forecasting accuracy, the purpose for ensureing QoS simultaneously, improve the resource utilization of container cluster.
What deserves to be explained is although foregoing teachings describe the essence of the invention by reference to several specific embodiments
God and principle, it should be appreciated that, the present invention is not limited to disclosed specific embodiment, the division also unawareness to various aspects
The feature that taste in these aspects cannot combine, this to divide merely to the convenience of statement.The present invention is directed to cover appended power
Included various modifications and equivalent arrangements in the spirit and scope of profit requirement.
Claims (4)
1. the Docker self-adapting dispatching systems that a kind of task perceives, which includes data collector, performance modeling device, response
Versus time estimator, feedforward controller, container scheduler and several containers,
Docker self-adapting dispatching method flows are：
1. data collector acquires the load of each container and CPU, memory system resource utilization；
2. performance modeling device is based on queueing theory, using step, 1. middle resource utilization as benchmark, builds application performance model, portrays
Load and the incidence relation of response time；
3. response time fallout predictor is estimated when being run by Kalman filter to performance model parameter, estimation is to meet
Prediction and actual measurement response error are expected to the condition of laying in a coffin；
4. feedforward controller analyzes residual error mean value and variance by fuzzy controller, before obtaining Kalman filter control parameter
Present adjusted value；
5. container scheduler judges whether the predicted value of response time has violated application service quality, and is carried out according to dispatching algorithm
Scheduling；
After 6. container scheduler performs container expansion, contraction or migration, step is continued to execute 1., forming method closed loop.
2. the Docker self-adapting dispatching systems that task as described in claim 1 perceives, which is characterized in that step 3. middle response
Versus time estimator uses adaptive Kalman filter to the prediction of response time, if formula is as follows：
X (k+1)=F (k) X (k)+Q (k), (3)
Z (k)=H (k) X (k)+R (k), (4)
Wherein, X (k) is prediction matrix, represents the matrix of service processing time and overall delay；
Z (k) is the state matrix of X (k)；
H (k) is responsible for the multi-C vector of observables being transformed into the multi-C vector of value to be predicted, is the resource profit by application example
With the matrix formed of rate, load and response time；
Q (k) is procedure activation noise covariance matrix,
R (k) is measurement noise covariance matrix,
Noise and procedure activation matrix are adjusted in real time by feedback control, to achieve the purpose that adaptive correction filtering parameter.
3. the Docker self-adapting dispatching systems that task as claimed in claim 2 perceives, which is characterized in that step 6. container tune
It spends device scheduling strategy and includes container migration, container expansion and containers shrink：
1) container migrates, and container transition process includes,
Container persistence is mirrored by controller first, preserves current application state, and then scheduler sends out traffic order, is received
Controller to traffic order from mirror image will generate 1 new container, then feed back to service registration mould according to traffic order
Block, to complete service discovery, original container will be deleted by controller, be equally also reported to and belonged to adaptive container resource confession
To the service register module of system, service deregistration is completed；
2) container expansion, the process of container expansion include, and scheduler is according to the report information of controller first, in current cluster
Suitable node is selected to send out and creates container order, and the configuration information of container is transferred to the controller in node, controller
The network address according to container and port numbers are subjected to service registration automatically after establishment new container；
3) containers shrink, the process of containers shrink include, and scheduler judges that the application is according to the report information of controller first
It is no to be shunk, and will determine that result feeds back to controller, controller carries out service deregistration automatically after receiving traffic order.
4. the adaptive container resource provision system that a kind of task perceives, is adaptively adjusted using Docker as claimed in claim 1
Degree method, which is characterized in that resource provision system includes automatic deployment module, resource flexible scheduling module, service registration and hair
Existing module,
Automatic deployment module is made of template warehouse and application deployment device, provided in template warehouse the deployment template that can refer to and
Application service, user selects concrete application and its version number according to demand for services, when template warehouse cannot meet user service need
When asking, user according to the autonomous drawing template establishment of template style,
Resource flexible scheduling module is made of the scheduler of main controlled node and the controller of child node, and the scheduler is according to user
Overall scheduling distribution is carried out with the scheduling request that the controller of child node transmits；
Service registration and discovery module realize that inside uses level tree-like storage using distributed consistency key assignments storage system
The process of structure, service registration and discovery is：When child node has new service creation, controller can be according to the configuration of the service
The network address of service and port numbers are registered to the service information store of main controlled node by information, memory Analysis Service
Configuration information classifies with being combined automatically to service,
The adaptive container resource provision working-flow that task perceives is as follows：
System uses host-guest architecture, and user is selected or created in template warehouse using required configuration, then by one group of configuration mould
Plate is transmitted to using deployment device, and configuration template is carried out using deployment device to be transmitted to application configuration information after fault-tolerant verification, combination
Global scheduler, global scheduler select suitable child node to be serviced according to the resource service condition of current each child node
Deployment, and by information on services persistent storage after the controller of child node receives traffic order, is specifically dispatched,
On the other hand, the information of controller also real-time collecting service is reported to the scheduler of main controlled node, when there is new demand servicing generation
When, service registration can be carried out from the memory of trend main controlled node.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711395307.6A CN108228347A (en) | 2017-12-21 | 2017-12-21 | The Docker self-adapting dispatching systems that a kind of task perceives |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711395307.6A CN108228347A (en) | 2017-12-21 | 2017-12-21 | The Docker self-adapting dispatching systems that a kind of task perceives |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108228347A true CN108228347A (en) | 2018-06-29 |
Family
ID=62647588
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711395307.6A Pending CN108228347A (en) | 2017-12-21 | 2017-12-21 | The Docker self-adapting dispatching systems that a kind of task perceives |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108228347A (en) |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108897627A (en) * | 2018-07-23 | 2018-11-27 | 南京叠嘉信息科技有限公司 | For the Docker dynamic dispatching algorithm of exemplary container |
| CN108984269A (en) * | 2018-07-16 | 2018-12-11 | 中山大学 | Container resource provision method and system based on random regression forest model |
| CN109151052A (en) * | 2018-09-19 | 2019-01-04 | 广州城市信息研究所有限公司 | Map Services register method, Map Services data capture method and device |
| CN109150603A (en) * | 2018-08-17 | 2019-01-04 | 中软信息系统工程有限公司 | The automatic expansion method of cloud server terminal and device |
| CN109150964A (en) * | 2018-07-06 | 2019-01-04 | 航天星图科技（北京）有限公司 | A kind of transportable data managing method and services migrating method |
| CN109618002A (en) * | 2019-01-11 | 2019-04-12 | 武汉斗鱼鱼乐网络科技有限公司 | A kind of micro services gateway optimization method, device and storage medium |
| CN109639791A (en) * | 2018-12-06 | 2019-04-16 | 广东石油化工学院 | Cloud workflow schedule method and system under a kind of container environment |
| CN109753356A (en) * | 2018-12-25 | 2019-05-14 | 北京友信科技有限公司 | A kind of container resource regulating method, device and computer readable storage medium |
| CN109886580A (en) * | 2019-02-26 | 2019-06-14 | 燕山大学 | A kind of intelligent plant control model and its management-control method |
| CN109945864A (en) * | 2019-02-25 | 2019-06-28 | 广州市香港科大霍英东研究院 | Indoor driving positioning fusion method, device, storage medium and terminal device |
| CN110022378A (en) * | 2019-04-22 | 2019-07-16 | 苏州排忧网络技术有限责任公司 | Vpn server smart allocation method based on state |
| CN110149396A (en) * | 2019-05-20 | 2019-08-20 | 华南理工大学 | A kind of platform of internet of things construction method based on micro services framework |
| CN110175068A (en) * | 2019-04-16 | 2019-08-27 | 平安科技（深圳）有限公司 | Host number elastic telescopic method, apparatus and computer equipment in distributed system |
| CN110288993A (en) * | 2019-06-26 | 2019-09-27 | 广州探迹科技有限公司 | A kind of individualized intelligent voice interactive method and device based on container technique |
| CN110764875A (en) * | 2019-10-25 | 2020-02-07 | 北京威晟艾德尔科技有限公司 | Docker container creating method based on competition mechanism |
| CN110837408A (en) * | 2019-09-16 | 2020-02-25 | 中国科学院软件研究所 | High-performance server-free computing method and system based on resource cache |
| CN110888741A (en) * | 2019-11-27 | 2020-03-17 | 新华三大数据技术有限公司 | Resource scheduling method and device for application container, server and storage medium |
| CN111045815A (en) * | 2018-11-29 | 2020-04-21 | 华为技术有限公司 | Method for optimizing deployed resources of multiple processors and expansion device thereof |
| CN111274111A (en) * | 2020-01-20 | 2020-06-12 | 西安交通大学 | Prediction and anti-aging method for microservice aging |
| CN111274576A (en) * | 2020-01-17 | 2020-06-12 | 济南浪潮高新科技投资发展有限公司 | Control method, system, equipment and medium for intelligent contract operating environment |
| CN111352726A (en) * | 2018-12-24 | 2020-06-30 | 北京奇虎科技有限公司 | Streaming data processing method and device based on containerized micro-service |
| CN111526179A (en) * | 2020-03-30 | 2020-08-11 | 中科九度(北京)空间信息技术有限责任公司 | Distributed node resource elastic expansion device and method |
| CN111782394A (en) * | 2020-06-29 | 2020-10-16 | 广东外语外贸大学 | Cluster service resource dynamic adjustment method based on response time perception |
| CN112181664A (en) * | 2020-10-15 | 2021-01-05 | 网易（杭州）网络有限公司 | Load balancing method and device, computer readable storage medium and electronic equipment |
| CN112685153A (en) * | 2020-12-25 | 2021-04-20 | 广州奇盾信息技术有限公司 | Micro-service scheduling method and device and electronic equipment |
| CN112783638A (en) * | 2019-11-07 | 2021-05-11 | 北京沃东天骏信息技术有限公司 | Method, apparatus, medium, and device for implementing data processing by calling service |
| CN112835714A (en) * | 2021-01-29 | 2021-05-25 | 中国人民解放军国防科技大学 | Container arrangement method, system and medium for CPU heterogeneous cluster in cloud edge environment |
| WO2021103790A1 (en) * | 2019-11-26 | 2021-06-03 | 北京京东尚科信息技术有限公司 | Container scheduling method and apparatus, and non-volatile computer-readable storage medium |
| CN113395178A (en) * | 2021-06-11 | 2021-09-14 | 聚好看科技股份有限公司 | Method and device for elastic expansion and contraction of container cloud |
| US11144362B1 (en) | 2020-05-05 | 2021-10-12 | International Business Machines Corporation | Container scheduling in a container orchestration system |
| CN113722108A (en) * | 2021-11-02 | 2021-11-30 | 中国工程物理研究院计算机应用研究所 | Container resource elastic pre-stretching method based on application operation load data prediction |
| WO2022016808A1 (en) * | 2020-07-22 | 2022-01-27 | 中国科学院深圳先进技术研究院 | Kubernetes cluster resource dynamic adjustment method and electronic device |
| CN114995900A (en) * | 2022-05-23 | 2022-09-02 | 中国联合网络通信集团有限公司 | Method and device for deploying multiple micro-service modules and computer readable storage medium |
| CN115190121A (en) * | 2022-05-25 | 2022-10-14 | 上海交通大学 | Cross-region-based micro-service overload scheduling system, method and equipment |
| US11748153B2 (en) | 2020-11-25 | 2023-09-05 | International Business Machines Corporation | Anticipated containerized infrastructure used in performing cloud migration |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105119913A (en) * | 2015-08-13 | 2015-12-02 | 东南大学 | Web server architecture based on Docker and interactive method between modules |
| CN105631196A (en) * | 2015-12-22 | 2016-06-01 | 中国科学院软件研究所 | Microservice-oriented container level flexible resource supply system and method |
| CN107426034A (en) * | 2017-08-18 | 2017-12-01 | 国网山东省电力公司信息通信公司 | A kind of extensive container scheduling system and method based on cloud platform |
-
2017
- 2017-12-21 CN CN201711395307.6A patent/CN108228347A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105119913A (en) * | 2015-08-13 | 2015-12-02 | 东南大学 | Web server architecture based on Docker and interactive method between modules |
| CN105631196A (en) * | 2015-12-22 | 2016-06-01 | 中国科学院软件研究所 | Microservice-oriented container level flexible resource supply system and method |
| CN107426034A (en) * | 2017-08-18 | 2017-12-01 | 国网山东省电力公司信息通信公司 | A kind of extensive container scheduling system and method based on cloud platform |
Non-Patent Citations (1)
| Title |
|---|
| 郝庭毅等: "面向微服务架构的容器级弹性资源供给方法", 《计算机研究与发展》 * |
Cited By (49)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109150964A (en) * | 2018-07-06 | 2019-01-04 | 航天星图科技（北京）有限公司 | A kind of transportable data managing method and services migrating method |
| CN109150964B (en) * | 2018-07-06 | 2022-01-11 | 中科星图股份有限公司 | Migratable data management method and service migration method |
| CN108984269A (en) * | 2018-07-16 | 2018-12-11 | 中山大学 | Container resource provision method and system based on random regression forest model |
| CN108984269B (en) * | 2018-07-16 | 2022-06-07 | 中山大学 | Container resource supply method and system based on random regression forest model |
| CN108897627A (en) * | 2018-07-23 | 2018-11-27 | 南京叠嘉信息科技有限公司 | For the Docker dynamic dispatching algorithm of exemplary container |
| CN108897627B (en) * | 2018-07-23 | 2021-11-09 | 南京叠嘉信息科技有限公司 | Docker dynamic scheduling method for typical container |
| CN109150603A (en) * | 2018-08-17 | 2019-01-04 | 中软信息系统工程有限公司 | The automatic expansion method of cloud server terminal and device |
| CN109151052A (en) * | 2018-09-19 | 2019-01-04 | 广州城市信息研究所有限公司 | Map Services register method, Map Services data capture method and device |
| CN109151052B (en) * | 2018-09-19 | 2022-04-19 | 广州城市信息研究所有限公司 | Map service registration method, map service data acquisition method and device |
| CN111045815A (en) * | 2018-11-29 | 2020-04-21 | 华为技术有限公司 | Method for optimizing deployed resources of multiple processors and expansion device thereof |
| CN111045815B (en) * | 2018-11-29 | 2024-04-09 | 华为云计算技术有限公司 | Method for optimizing deployed resources of multiple processors and extension equipment thereof |
| CN109639791A (en) * | 2018-12-06 | 2019-04-16 | 广东石油化工学院 | Cloud workflow schedule method and system under a kind of container environment |
| CN111352726B (en) * | 2018-12-24 | 2024-04-05 | 三六零科技集团有限公司 | Stream data processing method and device based on containerized micro-service |
| CN111352726A (en) * | 2018-12-24 | 2020-06-30 | 北京奇虎科技有限公司 | Streaming data processing method and device based on containerized micro-service |
| CN109753356A (en) * | 2018-12-25 | 2019-05-14 | 北京友信科技有限公司 | A kind of container resource regulating method, device and computer readable storage medium |
| CN109618002A (en) * | 2019-01-11 | 2019-04-12 | 武汉斗鱼鱼乐网络科技有限公司 | A kind of micro services gateway optimization method, device and storage medium |
| CN109945864A (en) * | 2019-02-25 | 2019-06-28 | 广州市香港科大霍英东研究院 | Indoor driving positioning fusion method, device, storage medium and terminal device |
| CN109886580A (en) * | 2019-02-26 | 2019-06-14 | 燕山大学 | A kind of intelligent plant control model and its management-control method |
| CN110175068A (en) * | 2019-04-16 | 2019-08-27 | 平安科技（深圳）有限公司 | Host number elastic telescopic method, apparatus and computer equipment in distributed system |
| CN110022378A (en) * | 2019-04-22 | 2019-07-16 | 苏州排忧网络技术有限责任公司 | Vpn server smart allocation method based on state |
| CN110022378B (en) * | 2019-04-22 | 2021-09-10 | 苏州排忧网络技术有限责任公司 | State-based VPN server intelligent distribution method |
| CN110149396B (en) * | 2019-05-20 | 2022-03-29 | 华南理工大学 | Internet of things platform construction method based on micro-service architecture |
| CN110149396A (en) * | 2019-05-20 | 2019-08-20 | 华南理工大学 | A kind of platform of internet of things construction method based on micro services framework |
| CN110288993A (en) * | 2019-06-26 | 2019-09-27 | 广州探迹科技有限公司 | A kind of individualized intelligent voice interactive method and device based on container technique |
| CN110837408A (en) * | 2019-09-16 | 2020-02-25 | 中国科学院软件研究所 | High-performance server-free computing method and system based on resource cache |
| CN110764875A (en) * | 2019-10-25 | 2020-02-07 | 北京威晟艾德尔科技有限公司 | Docker container creating method based on competition mechanism |
| CN112783638A (en) * | 2019-11-07 | 2021-05-11 | 北京沃东天骏信息技术有限公司 | Method, apparatus, medium, and device for implementing data processing by calling service |
| WO2021103790A1 (en) * | 2019-11-26 | 2021-06-03 | 北京京东尚科信息技术有限公司 | Container scheduling method and apparatus, and non-volatile computer-readable storage medium |
| CN110888741A (en) * | 2019-11-27 | 2020-03-17 | 新华三大数据技术有限公司 | Resource scheduling method and device for application container, server and storage medium |
| CN111274576A (en) * | 2020-01-17 | 2020-06-12 | 济南浪潮高新科技投资发展有限公司 | Control method, system, equipment and medium for intelligent contract operating environment |
| CN111274111A (en) * | 2020-01-20 | 2020-06-12 | 西安交通大学 | Prediction and anti-aging method for microservice aging |
| CN111526179A (en) * | 2020-03-30 | 2020-08-11 | 中科九度(北京)空间信息技术有限责任公司 | Distributed node resource elastic expansion device and method |
| US11144362B1 (en) | 2020-05-05 | 2021-10-12 | International Business Machines Corporation | Container scheduling in a container orchestration system |
| CN111782394B (en) * | 2020-06-29 | 2024-05-24 | 广东外语外贸大学 | Cluster service resource dynamic adjustment method based on response time perception |
| CN111782394A (en) * | 2020-06-29 | 2020-10-16 | 广东外语外贸大学 | Cluster service resource dynamic adjustment method based on response time perception |
| WO2022016808A1 (en) * | 2020-07-22 | 2022-01-27 | 中国科学院深圳先进技术研究院 | Kubernetes cluster resource dynamic adjustment method and electronic device |
| CN112181664B (en) * | 2020-10-15 | 2023-07-14 | 网易（杭州）网络有限公司 | Load balancing method and device, computer readable storage medium and electronic equipment |
| CN112181664A (en) * | 2020-10-15 | 2021-01-05 | 网易（杭州）网络有限公司 | Load balancing method and device, computer readable storage medium and electronic equipment |
| US11748153B2 (en) | 2020-11-25 | 2023-09-05 | International Business Machines Corporation | Anticipated containerized infrastructure used in performing cloud migration |
| CN112685153A (en) * | 2020-12-25 | 2021-04-20 | 广州奇盾信息技术有限公司 | Micro-service scheduling method and device and electronic equipment |
| CN112835714A (en) * | 2021-01-29 | 2021-05-25 | 中国人民解放军国防科技大学 | Container arrangement method, system and medium for CPU heterogeneous cluster in cloud edge environment |
| CN112835714B (en) * | 2021-01-29 | 2023-07-28 | 中国人民解放军国防科技大学 | Container arrangement method, system and medium for CPU heterogeneous clusters in cloud edge environment |
| WO2022257347A1 (en) * | 2021-06-11 | 2022-12-15 | 聚好看科技股份有限公司 | Container cloud autoscaling method, and cluster server |
| CN113395178A (en) * | 2021-06-11 | 2021-09-14 | 聚好看科技股份有限公司 | Method and device for elastic expansion and contraction of container cloud |
| CN113722108A (en) * | 2021-11-02 | 2021-11-30 | 中国工程物理研究院计算机应用研究所 | Container resource elastic pre-stretching method based on application operation load data prediction |
| CN114995900A (en) * | 2022-05-23 | 2022-09-02 | 中国联合网络通信集团有限公司 | Method and device for deploying multiple micro-service modules and computer readable storage medium |
| CN114995900B (en) * | 2022-05-23 | 2024-05-14 | 中国联合网络通信集团有限公司 | Method and device for deploying multiple micro-service modules and computer-readable storage medium |
| CN115190121A (en) * | 2022-05-25 | 2022-10-14 | 上海交通大学 | Cross-region-based micro-service overload scheduling system, method and equipment |
| CN115190121B (en) * | 2022-05-25 | 2023-11-17 | 上海交通大学 | Cross-region-based micro-service overload scheduling system, method and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108228347A (en) | The Docker self-adapting dispatching systems that a kind of task perceives | |
| Dhaya et al. | Energy-efficient resource allocation and migration in private cloud data centre | |
| US11146504B2 (en) | Market-based distributed resource allocation for edge-cloud systems | |
| Bhattacharjee et al. | Barista: Efficient and scalable serverless serving system for deep learning prediction services | |
| Téllez et al. | A tabu search method for load balancing in fog computing | |
| WO2021247448A1 (en) | Federated learning optimizations | |
| WO2019050952A1 (en) | Systems, methods, and media for distributing database queries across a metered virtual network | |
| WO2021026481A1 (en) | Methods, systems, articles of manufacture and apparatus to improve job scheduling efficiency | |
| EP3278578B1 (en) | Method and system for mtc event management | |
| Mechalikh et al. | PureEdgeSim: A simulation framework for performance evaluation of cloud, edge and mist computing environments | |
| Brogi et al. | Predictive analysis to support fog application deployment | |
| Alfakih et al. | Multi-objective accelerated particle swarm optimization with dynamic programing technique for resource allocation in mobile edge computing | |
| Nylander et al. | Cloud application predictability through integrated load-balancing and service time control | |
| Tsagkaropoulos et al. | Severity: a QoS-aware approach to cloud application elasticity | |
| Shefu et al. | Fruit fly optimization algorithm for network-aware web service composition in the cloud | |
| CN117596247A (en) | Resource monitoring and performance evaluation method based on heterogeneous edge computing system | |
| Foschini et al. | Adaptive and business-driven service placement in federated Cloud computing environments | |
| Reyes et al. | Adaptive multiscale optimization: Concept and case study on simulated UAV surveillance operations | |
| CN114978913B (en) | Cross-domain deployment method and system for service function chains based on cut chains | |
| De Grande et al. | Measuring communication delay for dynamic balancing strategies of distributed virtual simulations | |
| Štefanič et al. | Quality of Service‐aware matchmaking for adaptive microservice‐based applications | |
| De Vleeschauwer et al. | 5growth data-driven ai-based scaling | |
| CN115913967A (en) | Micro-service elastic scaling method based on resource demand prediction in cloud environment | |
| CN112948114B (en) | Edge computing method and edge computing platform | |
| CN114936089A (en) | Resource scheduling method, system, device and storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180629 |
|
| RJ01 | Rejection of invention patent application after publication |