CN105631196A - Microservice-oriented container level flexible resource supply system and method - Google Patents
Microservice-oriented container level flexible resource supply system and method Download PDFInfo
- Publication number
- CN105631196A CN105631196A CN201510974291.9A CN201510974291A CN105631196A CN 105631196 A CN105631196 A CN 105631196A CN 201510974291 A CN201510974291 A CN 201510974291A CN 105631196 A CN105631196 A CN 105631196A
- Authority
- CN
- China
- Prior art keywords
- response time
- container
- resource
- load
- equation
- 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
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a microservice-oriented container level flexible resource supply system and method. The system comprises a data acquisition device, a performance modeling device, a feedforward controller, a response time predictor and a container scheduler. According to the system and method, the deficiencies of difficulty in suitability for a sudden load mode and difficulty in user service quality guarantee in an existing method are made up for.
Description
Technical field
The present invention relates to a kind of container levels flexible resource feed system towards micro services framework and method, belong to the container resource elastic supply problem under the Internet and resource management field, particularly burst load scene.
Background technology
Micro services framework (Microservice) embodies the design philosophy of internet, applications, its core concept is that fine granularity Module Division, serviceization interface encapsulation, lightweight communication are mutual, there are following 2 advantages: (1) module autonomy is strong, internet, applications change can be met very well fast, the demand that module is independently updated, (2) module favorable expandability, can meet the prediction of internet, applications customer volume difficulty, the demand of Resource dynamic allocation very well. wherein, the former refers to application and development and design category, the latter, it is emphasised that application runs and maintenance issues, is also the emphasis paid close attention to herein. report according to Gartner, micro services itself has good autgmentability, it is increasingly becoming the main flow architecture mode of structure internet, applications, but from the visual angle run and safeguard, tackle typical the Internet sudden change load (Flash-crowds) scene, guarantee application service quality (QualityofService) still faces the challenge, and (service quality refers to the application software satisfaction degree to time requirement, response time is one of its important metric, such as QoS of customer is 5 seconds, namely represent that the interval that user initiates request response from request is not to be exceeded 5 seconds), such as Netflix, Facebook, Twitter etc.
In recent years, Lightweight Container technology is arisen at the historic moment, and its essence is simulation process running environment, has the features such as resource occupation is few, application startup is fast, just progressively becomes and support the main flow architecture platform that micro services is run. Container has the feature of second level resource provision, can well meet the demand that real time resources is supplied by internet, applications load changing. But, have method or be limited to physical machine motility not, resource is difficult to elastic supply; Or the supply minute level that is limited to resources of virtual machine is ageing, it is typically only capable to be applicable to the load model of mechanical periodicity (Time-of-day). Fast development and deep application along with interconnected technology, how to tackle sudden change load (Flash-crowds) still to face the challenge, as Jingdone district adopts container as " 618 " flash sale activity in 2015, user accesses number rate of increase higher than expection, causes that the phenomenon such as " without response ", " card " occurs in certain applications component.
Existing work is mainly directed towards physical machine and the elastic supply of virtual machine Scene realization resource. Some documents are difficult to the feature of quickly supply for physical resource, or adopt capacity planning, using service quality as constraints, estimate the peak resource demand of application; Or employing admission control mechanism, push away, according to resource provision amount is counter, the peak load that application can be born, ensure the service quality of application by refusing service strategy. As LudmilaCherkasova et al. propose based on admission control mechanism, set up the loss model of session and handling capacity, the demand and supply of deduction resource. The resource requirement equation group based on linear model that and for example RobertssonA et al. proposes, ensureing the lower in advance peak value estimating resource requirement of application quality, thus reaching the purpose of the on-demand supply of resource. Some documents consider virtual machine performance expense factor, adopt the method for model-driven to portray the resource requirement change of application under virtualized environment, and in this, as the foundation that resource provides. These methods generally adopt the enhancing study mechanism such as (ReinforcementLearning), statistical learning (StatisticalLearning) to carry out model parameter training. As KarlssonM et al. proposes the resource requirement of each Service Instance of methods analyst based on performance isolation, respectively to every kind of service construction performance change model, thus carrying out adaptive resource supply; Some documents consider reciprocity energy interference problem between virtual machine, adopt the methods such as machine learning (StatisticalMachineLearning), fuzzy control (FuzzyControl), theory of probability (ProbabilityTheory) to depict reciprocity between virtual machine and can disturb the impact of correspondence resource provision, and in this, as the foundation of application resource supply. AsThe mode based on machine learning that P et al. proposes, carries out rule by the performance parameter of historical data set correspondence model and trains, and carry out resource according to the resource provision rule obtained and dynamically adjust. But, owing to resources of virtual machine supply needs minute level, therefore said method is usually applicable only to the application scenarios of load cyclically-varying in time (Time-of-day).
Summary of the invention
The technology of the present invention solves problem: overcome the deficiencies in the prior art, it is provided that a kind of container levels resource provision method towards micro services framework, solves existing method and is difficult to be suitable for burst load pattern, there is the deficiency that QoS of customer is difficult to ensure.
The technology of the present invention solution: a kind of container levels flexible resource feed system towards micro services framework, including data acquisition unit, performance modeling device, response time predictor, feedforward controller and container scheduler, wherein:
Data acquisition unit, periodically monitoring container CPU, internal memory, magnetic disc i/o, network I/O system resource, and user's number of request per second; Additionally, data acquisition unit is responsible for setting up " load-preference resource utilization ", because the resource preference of container has diversity, such as I/O intensity container, the resource utilizations such as its disk transfers, network transmission are higher than the utilization rate of CPU possibly, and for example computation-intensive container, the utilization rate of its CPU and internal memory is likely to want more higher compared with disk I/O. Specifically, data acquisition unit periodically obtains the peak in cpu busy percentage, memory usage, magnetic disc i/o, these four parameters of network I/O, and is defined as preference resource utilization, then structure " load-preference resource utilization ";
Performance modeling device, according to the data that data acquisition unit obtains, is modeled by the performance equation in Jackson open loop networks and flow equation. Under preference resource utilization known conditions, build the tie-in equation of " load-response time " under micro services framework;
Response time predictor, there is known variables in " load-response time " tie-in equation, only determines this known variables, can calculate response time. Therefore, adopting Kalman filtering, using known variables as prediction matrix, using known variables as observing matrix, by known variables being carried out assignment, estimating response time, and estimated value and monitor value are obtained variance and average passes to feedforward controller;
Feedforward controller, analyze variance and the average of residual error, obtained the calibration value of filtering parameter (in " load-response time " tie-in equation the value of known variables) by fuzzy logic, its objective is to revise known variables so that the response time of estimation is more accurate;
Container scheduler, according to whether response time predictive value runs counter to application service quality, the on-demand supply carrying out container levels resource.
In described performance modeling device, build the incidence relation equation of " load-response time " under micro services framework by Jackson open loop networks specific as follows:
uj=u0j+��j����i(��ji��Tji)(1)
Wherein, j is application component, and i is application component example, j, and the relation of i can be described as: user's request will flow through multiple application component j1...jn, each application component j contains multiple example i1...im; Each example runs in a reservoir; uj�� [0,1) it is the preference resource utilization of application component j; u0jRefer to when application component j preference resource utilization in zero load situation; ��jiRefer to the number of concurrent of the i-th container of application component j, i.e. the number of request of arrival per second, meet poisson arrival process; TjiRefer to the service processing time of the i-th container of application component j; TjRefer to that the average service of application component j processes the time; D refers to service total delay time; B refers to service overall response time; ��jIt it is the load correlation coefficient with resource utilization of service j; uj,u0j,��ji, B obtains by monitoring, ��jIt it is the empirical value provided according to historical data;
Equation (1) represents preference resource utilization and adds overhead when having load equal to the expense of application build itself; Equation (2) represents response time equal to total delay plus total service processing time, and wherein, the service processing time is equal to the average service time of application build and the inverse ratio of untapped preference resource utilization; Such that it is able to obtain the performance equation of " load-response time " from " load-preference resource utilization ".
Described response time predictor detailed process.
(1) for equation (1) and equation (2), the average service process time of total delay and application component is difficult to predict, is unknown number; Preference resource utilization, application component own resources expense, load can be observed and obtaining, and is datum;
(2) by Kalman filtering algorithm, using unknown number as prediction matrix; Datum is as observing matrix;
(3) by carrying out assignment to unknown number, the predictive value of the time that meets with a response.
The filtering equations of Kalman is as follows:
X (k)=H (k) X (k-1)+Qk(3)
Z (k)=H (k) X (k)+Rk(4)
Wherein X (k) is prediction matrix, and its value isForm please adjust the matrix representing the service processing time with service flow total delay; Z (k) is observing matrix, and its value isForm please adjust the matrix representing the resource utilization of application example, load and response time; Q (k) is procedure activation noise covariance matrix, and it meets Q (k)��N (0, Q), R (k) and measures noise covariance matrix, and it meets the Gauss distribution of R (k)��N (0, R).
Described feedforward controller detailed process, in order to revise the unknown number in Kalman filtering, improves the accuracy of response time prediction.
(1) owing to burst load has uncertainty, it is therefore desirable to regulate the noise variance matrix of Kalman filter.
(2) noise variance matrix is consistent with the matrix of Gauss distribution, is represented by the first-order equation of standard deviation; And the residual error (difference of observing matrix actual value and predictive value) in wave filter calculating process, this difference represents the wave filter degree of dependence to measured value, when its value increases, representing load in burst change, the accuracy of filter prediction is declining, and this value is relevant with the standard deviation of noise variance matrix, so, can passing through to analyze variance and the average of residual error, adjusting the change of noise variance matrix, thus ensureing the accuracy of filter prediction.
Building Triangleshape grade of membership function by TS fuzzy logic, the curve of error of itself and conventional Kalman filter being carried out across comparison, thus inferring feasibility and the effectiveness of these group data. Again each experimental result and a front experimental result are carried out longitudinal contrast, to determine the linear combination of better effects if. Eventually pass 100 groups of emulation experiments, it is determined that the rule output data of FLAC, only enumerate two important FLAC rules here:
And if only if, and residual variance belongs to little, when residual error average belongs to zero, and T=P (r) �� 0.3+0.8, U=-P (r) �� 0.2+1.9
And if only if, and residual variance belongs to big, and residual error average belongs to hour, T=-P (r) �� 0.5+0.6, U=P (r) �� 0.1+1.4
Described container scheduler detailed process.
(1) whether response time runs counter to the standard that application service quality is container scheduling;
(2) when response time is broken a contract, to each application component service time response time shared by predictive value ratio in weight the higher person be extended; When response time is not broken a contract, if the total resources utilization rate of host is more than 75%, then the container that wherein resource occupation is higher is migrated; When response time is not broken a contract, if the resource utilization of the application example of certain application component is less than rated value, then shrink.
The container resource provision method of the present invention comprises the following steps:
Step S01: periodically monitoring gathers the system resources such as the CPU of each container, internal memory, magnetic disc i/o, network I/O and uses parameter and user's number of request per second; Each container is due to the difference of COS, and its resource preference is also different; Data acquisition phase generates " load-preference resource utilization " data pair the most at last, as its output;
Step S02: according to the data pair produced in step S01, be modeled by the performance equation in Jackson open loop networks and flow equation, portray the incidence relation of " load-response time " under micro services framework;
Step S03: using the predictive value as Kalman filter of the unknown number in " load-response time " relation equation in step S02, it is known that number, as the observations of Kalman filter, builds predictive equation and observational equation respectively;
Step S04: the residual error of predictive value and observation will be produced according in step S03 Kalman filter calculating process, residual error is to adjust the important parameter of Kalman filter prediction accuracy; Using the average of residual error and variance as the input of TS ambiguity function, the noise matrix of Kalman filter is output;
Step S05: the service quality whether having run counter to application according to the predictive value of the response time drawn in step S03 is foundation, container is scheduling, thus ensureing the service quality that user applies.
Present invention advantage compared with prior art is in that:
(1) present invention proposes the application elastic supply framework based on container, utilizes container lightweight feature, improves the actual effect of resource provision.
(2) present invention proposes the feature utilizing Fuzzy Adaptive Kalman Filtering Fast Convergent, improve the forecasting accuracy of sudden change load and the effectiveness of resource provision, solving existing method cannot under the scene of burst load, the problem ensureing QoS of customer.
Accompanying drawing explanation
Fig. 1 is the inventive method flowchart.
Fig. 2 is container scheduling strategy flow chart in the inventive method.
Detailed description of the invention
For making the present invention easier to understand, in conjunction with an example, the present invention is further elaborated, but this example does not constitute any limitation of the invention.
As shown in Figure 1: it is as follows that the present invention is embodied as step:
Step S01: the resource that when definition preference resource is that in CPU, internal memory, disk read-write, network service, container runs, proportion is of a relatively high; Data acquisition unit in the present embodiment, will be acquired the preference resource in each container and user's number of requests, and the data forming " preference resource-load " export as it; Preference resource refers to that each container is different due to COS, so the resource type of its preference is also different, the preference resource of computation-intensive is probably cpu busy percentage or memory usage, and preference resource intensive for I/O is probably magnetic disc i/o or network I/O.
Step S02: using the data of " preference resource-load " in step S01 to as input, by Jackson open loop networks performance equation and flow equation, portraying the incidence relation of load and response time;
Here, the relation of Jackson open loop networks and micro services framework refers to: in (1) micro services framework, application component is independent from, communicated by messaging bus between module, there is no status information, meet under Jackson network queuing model, node (application component and node) is separate, meets the constraint of exponential; (2) being interacted by message between application component under micro services framework, meeting Jackson network is open loop, and node input meets the hypothesis of Poisson distribution; (3) application component is after processing request, may select and enters next node or leave network.
User's request can redirect in node, through the process of related application component, finally responds to user. When certain application component exists multiple example time, adopt Round Robin (Round-RobinScheduling) strategy. In order to distinguish the different instances of same application component, definition: j is application component, and i is application component example. The relation of j, i three can be described as: user's request will flow through multiple application component j1...jn, each application component j contains multiple example i1...im; , each example runs in a reservoir. Due to the resource preference different (such as CPU collection type, I/O collection type) of application component, causing that the preference resource that container occurs is different, definition preference resource is the resource that in container CPU, internal memory, magnetic disc i/o, utilization rate is the highest; uj�� [0,1) it is the preference resource utilization of application component j; u0jRefer to when application component j preference resource utilization in zero load situation; ��jiRefer to the number of concurrent of the i-th container of application component j, i.e. the number of request of arrival per second, meet poisson arrival process; TjiRefer to the service processing time of the i-th container of application component j; TjRefer to that the average service of application component j processes the time; D refers to that user asks the overall network transmission time of stream f; B refers to the response time of service flow f; ��jIt it is the load correlation coefficient with resource utilization of service j. Have according to Jackson network traffics equation and network performance equation:
uj=u0j+��j����i(��ji��Tji)(1)
Wherein, uj,u0j,��ji, B obtains by monitoring, ��jIt is the empirical value provided according to historical data, Tji, d is difficult to monitor, it is necessary to estimate by predicting. So-called elastic supply, refers to that response time B is under the premise of relatively-stationary interval, the resource requirement of application. Visible, Tji, d is by the key element of elastic telescopic.
Step S03: using the predictive value as Kalman filter of the unknown number in " load-response time " relation equation in step S02, it is known that number, as the observations of Kalman filter, builds predictive equation and observational equation respectively;
Here, the original filtration equation of Kalman is as follows:
X (k)=H (k) X (k-1)+Qk(3)
Z (k)=H (k) X (k)+Rk(4)
Wherein X (k) is prediction matrix, and its value isRepresent the matrix of service processing time and service flow total delay; Z (k) is observing matrix, and its value isRepresent the matrix of the resource utilization of application example, load and response time; QkIt is procedure activation noise covariance matrix, RkBeing measure noise covariance matrix, it meets Rk��N (0, R) Gauss distribution, it is generally acknowledged and the two noise matrix should be set to zero-mean white noise, but load change is uncertain often, appearance such as load sudden peaks situation, so in order to make the scheduling of the flexible resource of system have a real-time, procedure activation noise covariance matrix and measure noise covariance matrix should adaptive adjustment in time, existing noise matrix is set to:
Qk=TQ (5)
Rk=UR (6)
Wherein, T, U is the adjusted value of time-varying, can obtain predictive equation as follows:
WhereinIt is the predictive value of each service processing time and request total delay time,It is the predictive value of resource utilization lamp observed parameter, it is possible to willPredictive value bring formula (1) and (2) into, draw the wave filter predictive value to response time.It is defined as residual error r, representative system model relies on the degree of measured value, the more big then system model of its value is more big to the dependence of measured value, at this moment illustrate that system load is likely to be to jump or the state of bust, service processing time and total delay cannot be carried out Accurate Prediction by wave filter, it is necessary to change Filtering Model.
Step S04: using the predictive value as Kalman filter of the unknown number in " load-response time " relation equation in step S02, it is known that number, as the observations of Kalman filter, builds predictive equation and observational equation respectively;
Judge wave filter the need of update according to exactly monitoring residual error, ideally residual error is zero-mean white noise, namely wave filter can perfect self adaptation, the average white noise if residual error is not zero, then illustrate that error occurs in filter prediction. Owing to residual variance, residual error average are relevant with Q and R, it is possible to by estimating residual variance and average, then carry out fuzzy reasoning, finally adjust the value of U and T, reach the purpose making Kalman filtering algorithm adapt to Time variable structure.
So, design an ambiguity function (FLAC) and constantly monitor the change of residual sum average, then adjust T and U according to fuzzy rule, to change noise matrix, thus the square error matrix of Kalman filter is adjusted so that it is perform optimal estimation, to meet time-dependent demand always.
Adopt TS fuzzy logic system, residual variance and average are set up Triangleshape grade of membership function and fuzzy rule. Such as, if residual variance is increasing, average also gradually away from zero, then should reduce procedure activation noise T and increase measurement noise U. Thus setting up fuzzy logic ordination table, zero in table represents T and U and need not change, and little representative increases T reduction U, and big representative reduces T and increases U, and middle representative increases T and U simultaneously.
Across comparison is carried out, thus inferring feasibility and the effectiveness of these group data according to above principle and by the curve of error of itself and conventional Kalman filter. Again each experimental result and a front experimental result are carried out longitudinal contrast, to determine the linear combination of better effects if. Eventually pass 100 groups of emulation experiments, it is determined that the rule output data of FLAC, only enumerate two important FLAC rules here:
And if only if, and residual variance belongs to little, when residual error average belongs to zero, and T=P (r) �� 0.3+0.8, U=-P (r) �� 0.2+1.9
And if only if, and residual variance belongs to big, and residual error average belongs to hour, T=-P (r) �� 0.5+0.6, U=P (r) �� 0.1+1.4
The effectiveness that the parameter of wave filter predicts the outcome dynamically is adjusted with guarantee according to above-mentioned FLAC rule.
Step S05: the service quality whether having run counter to application according to the predictive value of the response time drawn in step S03 is foundation, container is scheduling, thus ensureing the service quality that user applies;
According to load and resource service condition, container is carried out Real-Time Scheduling, export response time smoothly to ensure. Noise parameter in definition Kalman filtering algorithm is global variable T and U; In definition Service Instance, resource makes the maximum of consumption, minima and maximum response time; Definition Kalman filtering function EKF, for the predicated response time; Definition FLAC is ambiguity function, for T and U is sent feedforward information, the model parameter of Automatic adjusument EKF; Definition ResponseTime function, with the prediction output valve of EKF for parameter, calculates response time; Constant volume device migrates function Migrate, container expansion function Expand, containers shrink function Contract. First calculate response time predictive value and residual error average and variance according to Kalman filtering function, then bring the two value into fuzzy logic function, Kalman filtering function parameter is carried out feed-forward regulation, finally calculate the response time of application. Whether break a contract following three kinds of container scheduling strategies according to response time, seen Fig. 2:
(1) container migrates, and the main cause of its generation is that the use of container own resources reaches the resource limit upper limit, and the total resources being because host will arrive restriction upper limit threshold, now be not required to container is extended. Have only to migrate to partial containers other nodes. Container transition process approximately as, first container persistence is mirrored into, this step will preserve current application state, generates container according to mirror image on other nodes of cluster, and original container will be deleted by controller. When container migrates, the container that this strategy is minimum by migrating resource utilization, and judge that whether host resource utilization rate is less than threshold value, if being still above threshold value, then continue to migrate the container that resource utilization is minimum, until satisfying condition.
(2) container expansion, the main cause of its generation is in response to the time and runs counter to service quality, now carries out container and migrates and can not solve problem, for ensureing the average response time of this container, it is necessary to this container is extended. The process of container expansion approximately as, persistence is mirrored into by container, on other nodes according to this mirror image generate container, by the modes such as load balancing complete extension. During container expansion, by several container expansion higher for average service time to 2 times, all the other container instance numbers increase by one.
(3) containers shrink, the resource service condition that the main cause of its generation is each example of application is below estimated value, now needs the example quantity of this application of cutting. During containers shrink, will be less than the container instance number of threshold value and reduce one, if container instance number only one of which, then do not deal with.
Non-elaborated part of the present invention belongs to techniques well known.
Claims (6)
1. the container levels flexible resource feed system towards micro services framework, it is characterised in that including: data acquisition unit, performance modeling device, feedforward controller, response time predictor and container scheduler, wherein:
Data acquisition unit, periodically monitoring container CPU, internal memory, magnetic disc i/o, network I/O system resource, and user's number of request per second; It is also responsible for setting up " load-preference resource utilization ", the resource preference of container has diversity, data acquisition unit periodically obtains the peak in cpu busy percentage, memory usage, magnetic disc i/o, these four parameters of network I/O, and peak is defined as preference resource utilization, then structure " load-preference resource utilization ";
Performance modeling device, according to the data that data acquisition unit obtains, is modeled by the performance equation in Jackson open loop networks and flow equation, under preference resource utilization known conditions, builds the tie-in equation of " load-response time " under micro services framework;
Response time predictor, adopt Kalman filtering, using the known variables in the tie-in equation of " load-response time " under micro services framework as prediction matrix, using known variables as observing matrix, by known variables is carried out assignment, estimation response time, and variance and the average of estimated value and observation are passed to feedforward controller;
Feedforward controller, analyze variance and the average of residual error, obtained the calibration value of filtering parameter (in " load-response time " tie-in equation the value of known variables) by fuzzy logic, its objective is to revise known variables so that the response time of estimation is more accurate;
Container scheduler, according to whether response time predictive value runs counter to application service quality, the on-demand supply carrying out container levels resource.
2. the container levels flexible resource feed system towards micro services framework according to claim 1, it is characterised in that: in described performance modeling device, build the incidence relation equation of " load-response time " under micro services framework specific as follows:
uj=u0j+��j����i(��ji��Tji)(1)
Wherein, j is application component, and i is application component example, j, and the relation of i can be described as: user's request will flow through multiple application component j1...jn, each application component j contains multiple example i1...im; Each example runs in a reservoir; uj�� [0,1) it is the preference resource utilization of application component j; u0jRefer to when application component j preference resource utilization in zero load situation; ��jiRefer to the number of concurrent of the i-th container of application component j, i.e. the number of request of arrival per second, meet poisson arrival process; TjiRefer to the service processing time of the i-th container of application component j; TjRefer to that the average service of application component j processes the time; D refers to service total delay time; B refers to service overall response time; ��jIt it is the load correlation coefficient with resource utilization of service j; uj,u0j,��ji, B obtains by monitoring, ��jIt it is the empirical value provided according to historical data;
Equation (1) represents preference resource utilization and adds overhead when having load equal to the expense of application build itself; Equation (2) represents response time equal to total delay plus total service processing time, and wherein, the service processing time is equal to the average service time of application build and the inverse ratio of untapped preference resource utilization; Such that it is able to obtain the performance equation of " load-response time " from " load-preference resource utilization ".
3. the container levels flexible resource feed system towards micro services framework according to claim 1, it is characterised in that: the Kalman filtering side's device built in described response time predictor is as follows:
X (k)=H (k) X (k-1)+Qk(3)
Z (k)=H (k) X (k)+Rk(4)
Wherein X (k) is prediction matrix, represents the matrix of service processing time and service flow total delay; Z (k) is observing matrix, represents the matrix of the resource utilization of application example, load and response time; Q (k) is procedure activation noise covariance matrix, and R (k) measures noise covariance matrix.
4. the container levels flexible resource feed system towards micro services framework according to claim 1, it is characterized in that: the process of the calibration value that described feedforward fuzzy logic obtains Kalman filtering parameter is: build Triangleshape grade of membership function by TS fuzzy logic, the curve of error of curve of error and conventional Kalman filter is carried out across comparison, thus inferring feasibility and the effectiveness of these group data; Again each experimental result and a front experimental result are carried out longitudinal contrast, to determine the linear combination of better effects if; Eventually pass up to a hundred groups of emulation experiments, it is determined that rule output data, the i.e. calibration value of Kalman filtering parameter of FLAC.
5. the container levels flexible resource feed system towards micro services framework according to claim 1, it is characterised in that: described container scheduler detailed process is:
When the predictive value of response time is broken a contract, to each application component service time response time shared by predictive value ratio in weight the higher person be extended; When the predictive value of response time is not broken a contract, if the total resources utilization rate of host is more than 75%, then the container that wherein resource occupation is higher is migrated; When the predictive value of response time is not broken a contract, if the resource utilization of the application example of certain application component is less than rated value, then shrink.
6. the container levels flexible resource supply method towards micro services framework, it is characterised in that comprise the following steps:
Step S01: periodically gather the CPU of each container, internal memory, magnetic disc i/o, network I/O system resource use parameter and user's number of request per second; Each container is due to the difference of COS, and its resource preference is also different; Data acquisition phase generates " load-preference resource utilization " data pair the most at last;
Step S02: according to the data pair produced in step S01, be modeled by the performance equation in Jackson open loop networks and flow equation, builds the incidence relation of " load-response time " under micro services framework;
Step S03: using the predictive value as Kalman filter of the unknown number in " load-response time " relation equation in step S02, it is known that number, as the observations of Kalman filter, builds predictive equation and observational equation respectively;
Step S04: the residual error of predictive value and observation will be produced according in step S03 Kalman filter calculating process, residual error is to adjust the important parameter of Kalman filter prediction accuracy; Using the average of residual error and variance as the input of TS ambiguity function, the noise matrix of Kalman filter is output;
Step S05: whether the predictive value of the response time drawn in step S03 has run counter to the service quality of application is foundation, container is carried out resource management, thus ensureing the service quality that user applies.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510974291.9A CN105631196B (en) | 2015-12-22 | 2015-12-22 | A kind of container levels flexible resource feed system and method towards micro services framework |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510974291.9A CN105631196B (en) | 2015-12-22 | 2015-12-22 | A kind of container levels flexible resource feed system and method towards micro services framework |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105631196A true CN105631196A (en) | 2016-06-01 |
CN105631196B CN105631196B (en) | 2018-04-17 |
Family
ID=56046125
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510974291.9A Active CN105631196B (en) | 2015-12-22 | 2015-12-22 | A kind of container levels flexible resource feed system and method towards micro services framework |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105631196B (en) |
Cited By (36)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106227605A (en) * | 2016-07-26 | 2016-12-14 | 北京北森云计算股份有限公司 | The dynamic micro services expansion method of a kind of multilingual cloud compiling and device |
CN106330576A (en) * | 2016-11-18 | 2017-01-11 | 北京红马传媒文化发展有限公司 | Automatic scaling and migration scheduling method, system and device for containerization micro-service |
CN106533929A (en) * | 2016-12-30 | 2017-03-22 | 北京中电普华信息技术有限公司 | Micro-service development platform, generation method, deployment method and device |
CN107231291A (en) * | 2017-04-24 | 2017-10-03 | 全球能源互联网研究院 | A kind of micro services partition method and device suitable for electric network information physical system |
CN107992353A (en) * | 2017-07-31 | 2018-05-04 | 南京邮电大学 | A kind of container dynamic migration method and system based on minimum transition amount |
CN108206852A (en) * | 2016-12-20 | 2018-06-26 | 杭州华为数字技术有限公司 | A kind of dialogue-based Service Instance management method and equipment under micro services frame |
CN108228347A (en) * | 2017-12-21 | 2018-06-29 | 上海电机学院 | The Docker self-adapting dispatching systems that a kind of task perceives |
CN108234589A (en) * | 2016-12-22 | 2018-06-29 | 瞻博网络公司 | The auto zoom of micro services application |
CN108521352A (en) * | 2018-03-26 | 2018-09-11 | 天津大学 | Online cloud service tail delay prediction method based on stochastic reward net |
CN108664378A (en) * | 2018-05-10 | 2018-10-16 | 西安电子科技大学 | A kind of most short optimization method for executing the time of micro services |
CN108737224A (en) * | 2017-09-28 | 2018-11-02 | 新华三技术有限公司 | A kind of message processing method and device based on micro services framework |
CN108804154A (en) * | 2018-05-31 | 2018-11-13 | 天津大学 | A kind of automatic Synergistic methods of APP based on client micro services |
CN109167826A (en) * | 2018-08-20 | 2019-01-08 | 中软信息系统工程有限公司 | The restocking method, apparatus and system of WEB application |
CN109246167A (en) * | 2017-07-11 | 2019-01-18 | 阿里巴巴集团控股有限公司 | A kind of container dispatching method and device |
CN109417564A (en) * | 2016-07-22 | 2019-03-01 | 英特尔公司 | Technology for load of being assigned the job based on the utilization of resources stage |
CN109478146A (en) * | 2016-07-07 | 2019-03-15 | 思科技术公司 | System and method for application container of stretching in cloud environment |
CN109561134A (en) * | 2018-10-26 | 2019-04-02 | 平安科技(深圳)有限公司 | Electronic device, distributed type assemblies service distribution method and storage medium |
CN109617738A (en) * | 2018-12-28 | 2019-04-12 | 优刻得科技股份有限公司 | Method, system and the non-volatile memory medium of the scalable appearance of user service |
US10409664B2 (en) | 2017-07-27 | 2019-09-10 | International Business Machines Corporation | Optimized incident management using hierarchical clusters of metrics |
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 |
CN111130908A (en) * | 2019-12-31 | 2020-05-08 | 中信百信银行股份有限公司 | Micro-service dynamic aggregation and splitting system based on calling flow analysis and prediction |
CN111181773A (en) * | 2019-12-13 | 2020-05-19 | 西安交通大学 | Delay prediction method for multi-component application of heterogeneous border cloud collaborative intelligent system |
CN111274111A (en) * | 2020-01-20 | 2020-06-12 | 西安交通大学 | Prediction and anti-aging method for microservice aging |
CN111343219A (en) * | 2018-12-18 | 2020-06-26 | 同方威视技术股份有限公司 | Computing service cloud platform |
CN111431925A (en) * | 2020-04-02 | 2020-07-17 | 中国工商银行股份有限公司 | Message data processing method and device |
CN111522664A (en) * | 2020-04-24 | 2020-08-11 | 中国工商银行股份有限公司 | Service resource management and control method and device based on distributed service |
CN111611073A (en) * | 2020-05-08 | 2020-09-01 | 暨南大学 | Container placement method based on flow sensing under containerized data center |
CN111638959A (en) * | 2020-06-02 | 2020-09-08 | 山东汇贸电子口岸有限公司 | Elastic expansion method based on load regression prediction in cloud environment and readable storage medium |
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 |
WO2021073085A1 (en) * | 2019-10-14 | 2021-04-22 | 上海交通大学 | Microservice-oriented nanosecond-level power resource allocation method and system |
CN113064712A (en) * | 2021-04-16 | 2021-07-02 | 上海交通大学 | Micro-service optimization deployment control method, system and cluster based on cloud edge environment |
CN113450096A (en) * | 2021-06-25 | 2021-09-28 | 未鲲(上海)科技服务有限公司 | Resource transfer data processing method and device, electronic equipment and medium |
WO2022111156A1 (en) * | 2020-11-24 | 2022-06-02 | International Business Machines Corporation | Automated orchestration of containers by assessing microservices |
CN115022311A (en) * | 2022-03-23 | 2022-09-06 | 北京邮电大学 | Selection method and device of micro-service container instances |
CN116627660A (en) * | 2023-07-24 | 2023-08-22 | 湖北省楚天云有限公司 | Micro-service resource allocation method based on cloud data center |
CN116932233A (en) * | 2023-09-19 | 2023-10-24 | 金网络(北京)数字科技有限公司 | Micro-service architecture of intelligent contract |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101382912A (en) * | 2008-09-02 | 2009-03-11 | 中国科学院软件研究所 | Method for optimizing resource sensitive property orienting application server and system thereof |
CN102130938A (en) * | 2010-12-03 | 2011-07-20 | 中国科学院软件研究所 | Resource supply method oriented to Web application host platform |
US20120291088A1 (en) * | 2011-05-10 | 2012-11-15 | Sybase, Inc. | Elastic resource provisioning in an asymmetric cluster environment |
CN103064820A (en) * | 2012-12-26 | 2013-04-24 | 无锡江南计算技术研究所 | Cluster calculating system based on reconfigurable micro-server |
CN103076849A (en) * | 2012-12-26 | 2013-05-01 | 无锡江南计算技术研究所 | Reconfigurable micro server system |
CN105162884A (en) * | 2015-09-25 | 2015-12-16 | 浪潮(北京)电子信息产业有限公司 | Cloud management platform based on micro-service architecture |
-
2015
- 2015-12-22 CN CN201510974291.9A patent/CN105631196B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101382912A (en) * | 2008-09-02 | 2009-03-11 | 中国科学院软件研究所 | Method for optimizing resource sensitive property orienting application server and system thereof |
CN102130938A (en) * | 2010-12-03 | 2011-07-20 | 中国科学院软件研究所 | Resource supply method oriented to Web application host platform |
US20120291088A1 (en) * | 2011-05-10 | 2012-11-15 | Sybase, Inc. | Elastic resource provisioning in an asymmetric cluster environment |
CN103064820A (en) * | 2012-12-26 | 2013-04-24 | 无锡江南计算技术研究所 | Cluster calculating system based on reconfigurable micro-server |
CN103076849A (en) * | 2012-12-26 | 2013-05-01 | 无锡江南计算技术研究所 | Reconfigurable micro server system |
CN105162884A (en) * | 2015-09-25 | 2015-12-16 | 浪潮(北京)电子信息产业有限公司 | Cloud management platform based on micro-service architecture |
Non-Patent Citations (1)
Title |
---|
郭栋 等: "一种基于微服务架构的新型云件", 《技术研究》 * |
Cited By (61)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109478146A (en) * | 2016-07-07 | 2019-03-15 | 思科技术公司 | System and method for application container of stretching in cloud environment |
CN109417564A (en) * | 2016-07-22 | 2019-03-01 | 英特尔公司 | Technology for load of being assigned the job based on the utilization of resources stage |
CN109417564B (en) * | 2016-07-22 | 2022-04-15 | 英特尔公司 | Coordinator server, method and medium thereof |
CN106227605A (en) * | 2016-07-26 | 2016-12-14 | 北京北森云计算股份有限公司 | The dynamic micro services expansion method of a kind of multilingual cloud compiling and device |
CN106227605B (en) * | 2016-07-26 | 2019-10-18 | 北京北森云计算股份有限公司 | A kind of the dynamic micro services expansion method and device of multilingual cloud compiling |
CN106330576A (en) * | 2016-11-18 | 2017-01-11 | 北京红马传媒文化发展有限公司 | Automatic scaling and migration scheduling method, system and device for containerization micro-service |
CN106330576B (en) * | 2016-11-18 | 2019-10-25 | 北京红马传媒文化发展有限公司 | The method of containerization micro services automatic telescopic and migration scheduling, system and equipment |
CN108206852A (en) * | 2016-12-20 | 2018-06-26 | 杭州华为数字技术有限公司 | A kind of dialogue-based Service Instance management method and equipment under micro services frame |
CN108206852B (en) * | 2016-12-20 | 2020-12-22 | 华为技术有限公司 | Session-based service instance management method and device under micro-service framework |
CN108234589A (en) * | 2016-12-22 | 2018-06-29 | 瞻博网络公司 | The auto zoom of micro services application |
CN106533929A (en) * | 2016-12-30 | 2017-03-22 | 北京中电普华信息技术有限公司 | Micro-service development platform, generation method, deployment method and device |
CN106533929B (en) * | 2016-12-30 | 2019-08-23 | 北京中电普华信息技术有限公司 | A kind of micro services development system, generation method and dispositions method and device |
CN107231291B (en) * | 2017-04-24 | 2022-04-05 | 全球能源互联网研究院 | Micro-service isolation method and device suitable for power grid information physical system |
CN107231291A (en) * | 2017-04-24 | 2017-10-03 | 全球能源互联网研究院 | A kind of micro services partition method and device suitable for electric network information physical system |
CN109246167A (en) * | 2017-07-11 | 2019-01-18 | 阿里巴巴集团控股有限公司 | A kind of container dispatching method and device |
CN109246167B (en) * | 2017-07-11 | 2022-04-01 | 阿里巴巴集团控股有限公司 | Container scheduling method and device |
US11163631B2 (en) | 2017-07-27 | 2021-11-02 | International Business Machines Corporation | Optimized incident management using hierarchical clusters of metrics |
US10409664B2 (en) | 2017-07-27 | 2019-09-10 | International Business Machines Corporation | Optimized incident management using hierarchical clusters of metrics |
CN107992353B (en) * | 2017-07-31 | 2022-03-18 | 南京邮电大学 | Container dynamic migration method and system based on minimum migration volume |
CN107992353A (en) * | 2017-07-31 | 2018-05-04 | 南京邮电大学 | A kind of container dynamic migration method and system based on minimum transition amount |
CN108737224A (en) * | 2017-09-28 | 2018-11-02 | 新华三技术有限公司 | A kind of message processing method and device based on micro services framework |
CN108737224B (en) * | 2017-09-28 | 2020-06-09 | 新华三技术有限公司 | Message processing method and device based on micro-service architecture |
CN108228347A (en) * | 2017-12-21 | 2018-06-29 | 上海电机学院 | The Docker self-adapting dispatching systems that a kind of task perceives |
CN108521352A (en) * | 2018-03-26 | 2018-09-11 | 天津大学 | Online cloud service tail delay prediction method based on stochastic reward net |
CN108521352B (en) * | 2018-03-26 | 2022-07-22 | 天津大学 | Online cloud service tail delay prediction method based on random return network |
CN108664378B (en) * | 2018-05-10 | 2020-04-14 | 西安电子科技大学 | Method for optimizing minimum execution time of micro-service |
CN108664378A (en) * | 2018-05-10 | 2018-10-16 | 西安电子科技大学 | A kind of most short optimization method for executing the time of micro services |
CN108804154A (en) * | 2018-05-31 | 2018-11-13 | 天津大学 | A kind of automatic Synergistic methods of APP based on client micro services |
CN109167826B (en) * | 2018-08-20 | 2021-05-07 | 中软信息系统工程有限公司 | Method, device and system for putting WEB application on shelf |
CN109167826A (en) * | 2018-08-20 | 2019-01-08 | 中软信息系统工程有限公司 | The restocking method, apparatus and system of WEB application |
CN109561134B (en) * | 2018-10-26 | 2022-06-10 | 平安科技(深圳)有限公司 | Electronic device, distributed cluster service distribution method and storage medium |
CN109561134A (en) * | 2018-10-26 | 2019-04-02 | 平安科技(深圳)有限公司 | Electronic device, distributed type assemblies service distribution method and storage medium |
CN111343219A (en) * | 2018-12-18 | 2020-06-26 | 同方威视技术股份有限公司 | Computing service cloud platform |
CN109617738A (en) * | 2018-12-28 | 2019-04-12 | 优刻得科技股份有限公司 | Method, system and the non-volatile memory medium of the scalable appearance of user service |
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 |
WO2021073085A1 (en) * | 2019-10-14 | 2021-04-22 | 上海交通大学 | Microservice-oriented nanosecond-level power resource allocation method and system |
CN111181773A (en) * | 2019-12-13 | 2020-05-19 | 西安交通大学 | Delay prediction method for multi-component application of heterogeneous border cloud collaborative intelligent system |
CN111181773B (en) * | 2019-12-13 | 2021-09-07 | 西安交通大学 | Delay prediction method for multi-component application of heterogeneous border cloud collaborative intelligent system |
CN111130908B (en) * | 2019-12-31 | 2022-09-06 | 中信百信银行股份有限公司 | Micro-service dynamic aggregation and splitting system based on calling flow analysis and prediction |
CN111130908A (en) * | 2019-12-31 | 2020-05-08 | 中信百信银行股份有限公司 | Micro-service dynamic aggregation and splitting system based on calling flow analysis and prediction |
CN111274111B (en) * | 2020-01-20 | 2021-11-19 | 西安交通大学 | Prediction and anti-aging method for microservice aging |
CN111274111A (en) * | 2020-01-20 | 2020-06-12 | 西安交通大学 | Prediction and anti-aging method for microservice aging |
CN111431925A (en) * | 2020-04-02 | 2020-07-17 | 中国工商银行股份有限公司 | Message data processing method and device |
CN111431925B (en) * | 2020-04-02 | 2022-08-26 | 中国工商银行股份有限公司 | Message data processing method and device |
CN111522664A (en) * | 2020-04-24 | 2020-08-11 | 中国工商银行股份有限公司 | Service resource management and control method and device based on distributed service |
CN111611073A (en) * | 2020-05-08 | 2020-09-01 | 暨南大学 | Container placement method based on flow sensing under containerized data center |
CN111611073B (en) * | 2020-05-08 | 2023-01-31 | 暨南大学 | Container placement method based on flow sensing in containerized data center |
CN111638959A (en) * | 2020-06-02 | 2020-09-08 | 山东汇贸电子口岸有限公司 | Elastic expansion method based on load regression prediction in cloud environment and readable storage medium |
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 |
WO2022111156A1 (en) * | 2020-11-24 | 2022-06-02 | International Business Machines Corporation | Automated orchestration of containers by assessing microservices |
US11704123B2 (en) | 2020-11-24 | 2023-07-18 | International Business Machines Corporation | Automated orchestration of containers by assessing microservices |
GB2615040A (en) * | 2020-11-24 | 2023-07-26 | Ibm | Automated orchestration of containers by assessing microservices |
CN113064712A (en) * | 2021-04-16 | 2021-07-02 | 上海交通大学 | Micro-service optimization deployment control method, system and cluster based on cloud edge environment |
CN113064712B (en) * | 2021-04-16 | 2022-05-17 | 上海交通大学 | Micro-service optimization deployment control method, system and cluster based on cloud edge environment |
CN113450096A (en) * | 2021-06-25 | 2021-09-28 | 未鲲(上海)科技服务有限公司 | Resource transfer data processing method and device, electronic equipment and medium |
CN115022311A (en) * | 2022-03-23 | 2022-09-06 | 北京邮电大学 | Selection method and device of micro-service container instances |
CN115022311B (en) * | 2022-03-23 | 2023-06-09 | 北京邮电大学 | Method and device for selecting micro-service container instance |
CN116627660A (en) * | 2023-07-24 | 2023-08-22 | 湖北省楚天云有限公司 | Micro-service resource allocation method based on cloud data center |
CN116932233A (en) * | 2023-09-19 | 2023-10-24 | 金网络(北京)数字科技有限公司 | Micro-service architecture of intelligent contract |
CN116932233B (en) * | 2023-09-19 | 2023-12-08 | 金网络(北京)数字科技有限公司 | Micro-service architecture of intelligent contract |
Also Published As
Publication number | Publication date |
---|---|
CN105631196B (en) | 2018-04-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105631196A (en) | Microservice-oriented container level flexible resource supply system and method | |
Toka et al. | Machine learning-based scaling management for kubernetes edge clusters | |
US9386086B2 (en) | Dynamic scaling for multi-tiered distributed systems using payoff optimization of application classes | |
CN111953758B (en) | Edge network computing unloading and task migration method and device | |
Qi et al. | Scalable parallel task scheduling for autonomous driving using multi-task deep reinforcement learning | |
Zhou et al. | Reinforcement learning-based adaptive resource management of differentiated services in geo-distributed data centers | |
Filali et al. | Preemptive SDN load balancing with machine learning for delay sensitive applications | |
Kim et al. | Multi-agent reinforcement learning-based resource management for end-to-end network slicing | |
Duggan et al. | A multitime‐steps‐ahead prediction approach for scheduling live migration in cloud data centers | |
CN104468413A (en) | Network service method and system | |
Zhang et al. | Deep reinforcement learning based cooperative partial task offloading and resource allocation for IIoT applications | |
CN107483292A (en) | Dynamic monitoring and controlling method for cloud platform | |
CN109102155A (en) | A kind of ultra-short term deploying node probability forecasting method and system | |
CN113407249A (en) | Task unloading method facing to position privacy protection | |
CN113641445B (en) | Cloud resource self-adaptive configuration method and system based on depth deterministic strategy | |
Mazidi et al. | An autonomic decision tree‐based and deadline‐constraint resource provisioning in cloud applications | |
Xiao et al. | Dscaler: A horizontal autoscaler of microservice based on deep reinforcement learning | |
Tao et al. | DRL-Driven Digital Twin Function Virtualization for Adaptive Service Response in 6G Networks | |
Chen et al. | Data center optimal regulation service reserve provision with explicit modeling of quality of service dynamics | |
Yan et al. | Service caching for meteorological emergency decision-making in cloud-edge computing | |
Li et al. | Efficient data offloading using markovian decision on state reward action in edge computing | |
Sun et al. | Performance analysis of mobile cloud computing with bursty demand: A tandem queue model | |
CN115858153A (en) | Cloud computing platform resource management system facing smart city rail and dynamic allocation method | |
CN115913967A (en) | Micro-service elastic scaling method based on resource demand prediction in cloud environment | |
Liang et al. | Stochastic Stackelberg Game Based Edge Service Selection for Massive IoT Networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |