CN107395733B - Geographic distribution interactive service cloud resource collaborative optimization method - Google Patents

Geographic distribution interactive service cloud resource collaborative optimization method Download PDF

Info

Publication number
CN107395733B
CN107395733B CN201710639715.5A CN201710639715A CN107395733B CN 107395733 B CN107395733 B CN 107395733B CN 201710639715 A CN201710639715 A CN 201710639715A CN 107395733 B CN107395733 B CN 107395733B
Authority
CN
China
Prior art keywords
resource
delay
service
data center
resources
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.)
Active
Application number
CN201710639715.5A
Other languages
Chinese (zh)
Other versions
CN107395733A (en
Inventor
姚建国
吴家宏
管海兵
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Jiaotong University
Original Assignee
Shanghai Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CN201710639715.5A priority Critical patent/CN107395733B/en
Publication of CN107395733A publication Critical patent/CN107395733A/en
Priority to PCT/CN2018/072031 priority patent/WO2019024445A1/en
Application granted granted Critical
Publication of CN107395733B publication Critical patent/CN107395733B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0206—Price or cost determination based on market factors
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/51—Discovery or management thereof, e.g. service location protocol [SLP] or web services
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/60—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Accounting & Taxation (AREA)
  • Signal Processing (AREA)
  • Development Economics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Data Mining & Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a collaborative optimization method for cloud resources of a geographically distributed interactive service, which comprises the following steps: step 1: determining available data centers for each area; step 2: determining that the interactive service can schedule and distribute all possible data center combinations; and step 3: designing and appointing a data center resource providing mode; and 4, step 4: determining a price demand probability distribution model corresponding to a stochastic programming algorithm; and 5: determining a price model according to the resource providing plan formulated in the step 3; step 6: defining tail delay of interactive service, and using the tail delay as delay constraint of optimization model; and 7: determining a data center resource scheduling optimization model; and 8: decomposing the overall problem into a main problem and a series of sub-problems by using the characteristics of random programming; and step 9: and (6) solving an optimization algorithm. The present invention aims to minimize resource configuration costs for geographically distributed interactive services by a multi-cloud service provider.

Description

Geographic distribution interactive service cloud resource collaborative optimization method
Technical Field
The invention relates to an optimization method, in particular to a collaborative optimization method for cloud resources of a geographically distributed interactive service.
Background
Geographically distributed interactive services are a class of delay sensitive computing services that need to be deployed to multiple data center areas (e.g., data center areas such as north america, asia-pacific, and europe). For interactive services such as Web search and real-time data analysis, a user sends a Web request to a data center of service deployment to acquire data or perform decision analysis. Since interactive services rely on the basic data provided by multiple data center areas, which are distributed globally, it is decided that geographically distributed interactive services require the selection of a suitable set of data centers for deployment, and that the set of data centers consists of at least one data center provided by each data center area. In addition, in order to meet the deployment requirement of the interactive service, the data center is required to provide sufficient resources to run the data center workload supporting the interactive service. Therefore, due to the dependence of basic data and the requirement of data center resources, the deployment position of the geographic distribution interactive service and the data center resource plan need to meet the tail delay requirement of users. The ultimate goal is to minimize the overall computational and communication costs of geographically distributed interactive services.
Today's cloud computing services are already well established. A certain number of data centers are deployed in regions on the global scale, such as Google computer Engine, Amazon AWS, Microsoft Azure, Ariicloud and the like, and all the data centers can provide complete cloud resource use schemes. Google computer Engine and Microsoft Azure provide policies for using data center resources on demand. They provide VM instances of differentiated computing and storage capabilities and corresponding prices. In particular, the actual usage cost of a resource is calculated in terms of usage time, e.g., a user rents a VM instance for 10 hours, then the total cost is the unit price of the VM instance multiplied by 10 hours. Amazon AWS not only provides a policy like Google computerengine and Microsoft Azure for on demand use of resources, but Amazon AWS also allows users to make a reservation resource contract, where users can reserve 1 or 3 year specifically configured VM instances and pay resource reservation fees in advance, and then bill for actual usage time at a lower price (around 50% of the price of on demand resources). On one hand, the resource prices of the cloud service providers are different, on the other hand, the resource prices of the data centers in different areas of the same cloud service provider are also different, and the resource prices are continuously changed according to market demands. Because the interactive service requests to access the data center to obtain the computing service through the Web request, the cloud service provider rates the use of the WAN network bandwidth according to the total amount of the used traffic, and the unit price of the traffic is generally in a ladder shape. For example, AWS is free for the first 1GB, more than 1G less than 10TB is charged for $0.090/GB, and more than 10TB less than 40TB is charged for $ 0.085/GB.
Since geographically distributed interactive services are a class of delay sensitive Web services, an appropriate amount of data center resources are required to provide the quality of service desired by the user for different service request rates. In the statistical information of the request rates of the Google computer Engine for the interactive service in different periods, the difference of the service request rates is great, and the difference between the lowest request rate and the highest request is 103More than twice. Therefore, the uncertainty of the resource plan adapting to the requirement for the deployment of the interactive service is an urgent solutionTo a problem of (a). Although current cloud service providers offer policies based on reservation or on-demand use of resources, they all expose some drawbacks. If the user uses the resources based on reservation, in order to guarantee the service quality, the user must use the reserved large amount of resources to guarantee that the interactive service can still meet the service quality requirement of the user when the request rate is maximum. But because of the uncertainty and large differences in the demand, this results in a large amount of wasted resources, which is expensive and often does not work. When the user uses only the on-demand resource, although the resource demand of the service can be dynamically satisfied, since the price of the on-demand resource is higher than the price of the reserved resource by more than 50%, the resource cost is not much different from that of the strategy of utilizing the reserved resource. Based on the premise, a strategy capable of mixing reserved resources and resources on demand needs to be designed, so that the deployment cost of the geographically distributed interactive service is saved.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a geographical distribution interactive service cloud resource collaborative optimization method, which aims to minimize the resource configuration cost of the geographical distribution interactive service through a multi-cloud service provider.
According to one aspect of the invention, a method for collaborative optimization of cloud resources of geographically distributed interactive services is provided, which is characterized by comprising the following steps:
step 1: determining available data centers for each area;
step 2: determining all available data center combinations that the interactive service can mobilize;
and step 3: designing and appointing a data center resource providing mode;
and 4, step 4: determining a price demand probability distribution model corresponding to a stochastic programming algorithm;
and 5: determining a price model according to the resource providing plan formulated in the step 3;
step 6: defining tail delay of the interactive service and using the tail delay as a delay constraint of an optimization model;
and 7: determining a data center resource scheduling optimization model;
decomposing the overall problem into a main problem and a series of sub-problems by using the characteristics of random programming;
and 8: solving an optimization algorithm, recording the resource demand of each interactive service and corresponding delay information, determining the relation between the resource quantity and the delay and predicting the primary workload scheduling; when the service is on-line, the resource demand and the delay are correspondingly changed, and the changes are used for random planning and finally applied to the dynamic adjustment of the resource plan.
Preferably, the step 3 specifically includes the following steps:
step 3.1: using reserved resource pattern only; the data center provides reserved resources, and the cloud consumer firstly makes a reserved contract and pays a prepaid fee according to the service deployment cycle and the compromised resource requirements; after the interactive service is on line, firstly, the reserved resources are distributed according to the actual requirements of the service resources, and if the reserved resources can meet the requirements of the service quality in the deployment period, the service quality requirements of the interactive service can be met only by reserving the resources;
step 3.2: using only on-demand resource patterns; the actual demand of the service is greater than the maximum available resource number of the lease due to the change of the request rate, so that the required service quality can not be achieved, and extra resources on demand are leased on the basis of reserved resources to achieve the requirement of the service quality, wherein the situation is called a hybrid phase;
step 3.3: in the mixed mode, the cloud consumers perform short-term deployment of services according to the resources on demand, and the data center only provides the resources on demand.
Preferably, the step 5 specifically includes the following steps:
step 5.1: determining a pre-paid price of the reserved contract;
step 5.2: determining a dynamic price of an actual running VM instance;
step 5.3: determining a network resource charging mode and price;
step 5.4: overlapping the cost of the computing resources and the cost of the network resources to obtain a total price;
step 5.5: a dynamic cost of the instance running in real time is determined.
Preferably, the step 6 specifically includes the following steps:
step 6.1: determining a calculation delay and a network transmission delay;
step 6.2: determining a delay threshold value of each request, and establishing a delay threshold value matrix;
step 6.3: and (4) utilizing a service level agreement to measure whether the tail delay meets the service quality requirement under the condition of determining the resource providing plan.
Preferably, said step 6.3 comprises in particular the steps of:
and 6.3.1, performing probability analysis on the historical data by using a probability statistical method to obtain a preliminary relation between the resource providing plan and the S L A tail delay.
Preferably, said step 6.3.1 comprises in particular the steps of:
step 6.3.1.1: calculating delay and network transmission delay probability statistical analysis;
step 6.3.1.2: calculating a probabilistic delay for each request;
step 6.3.1.3: and calculating probability tail delay.
Preferably, said step 7 comprises the steps of:
step 7.1: calculating the deployment cost of the candidate data center of each area;
step 7.2: and determining a main random planning problem and a sub-random planning problem.
Preferably, said step 8 comprises the steps of:
step 8.1: scheduling the workload;
step 8.2: the workload resource plan is adjusted in real time.
Preferably, said step 8.1 comprises the steps of:
step 8.1.1 firstly, generating probability distribution of VCPU (virtual central processing unit) requirements, memory requirements, disk requirements and network utilization rate according to the request record of the geographic distribution interactive service;
step 8.1.2, the generated probability distribution is used as configuration data, and the corresponding deployment cost of each schedulable data center is obtained;
step 8.1.3 initializes the scheduling decision variables.
Preferably, said step 8.2 comprises the steps of:
step 8.2.1, through the scheduling of the workload, obtaining target data center groups of all interactive services, but the obtained preliminary resource plan is coarse-grained, and the resource plan needs to be adjusted at any RPP time, so that corresponding resource demand probability distribution is generated at each RPP time;
step 8.2.2, solving by taking the generated probability distribution as configuration data to obtain a resource plan of each working load in the target data center, wherein the resource plan is fine-grained;
step 8.2.3 selects a target data center group for the geographically distributed interactive service per WDD time and reconfigures the resource plan at each RPP time, where each WDD time includes a plurality of RPP times.
Compared with the prior art, the method has the advantages that the method aims to minimize the resource allocation cost of the geographic distribution interactive service through multiple cloud service providers, because of the uncertainty of the resource price of the data center and the characteristics of the dynamic resource demand of the interactive service, a price model and an S L A delay model are designed, the scheduling of the workload and the resource plan are respectively described as a main method and a series of sub-problems based on the obtained constraint conditions and combined with random planning, the method is the essence of the method, the optimized design and the constraint method obtained through price modeling and S L A delay modeling have better universality and expansibility for the resource scheduling of most of the geographic distribution interactive service, the method saves the cost by 24% compared with the service deployment based on the resource demand, the cost by 10% compared with the service deployment based on the reserved resource, the delayed S L A constraint, the tail delay constraint based on negotiation can be met, the tail delay constraint based on the consideration can be met, the research aiming at reducing the service deployment cost of the global distribution is also lacked, and the research on the cost of some related research on the application of the global distribution of the data center is created and the corresponding optimization algorithm is created.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a schematic diagram of a VCPU demand probability distribution.
FIG. 2 is a schematic diagram of memory demand probability distribution.
FIG. 3 is a schematic diagram of a disk demand probability distribution.
Fig. 4 is a schematic diagram of a network usage probability distribution.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications can be made by persons skilled in the art without departing from the spirit of the invention. All falling within the scope of the present invention.
Geographically distributed interactive services require support from the underlying data, which is typically backed up between data centers in the same area in order to provide high availability. Because of the backup of the basic data, the interactive service can select a data center with lower price for deployment. However, the prices of the various cloud service providers are different for the same configured VM instance and the same network traffic. In addition, the pressure of requests for services by users of interactive services at different times varies, thus making the demand for resources by interactive services uncertain. Because of changing resource demands and changing data center resource prices, geographically distributed interactive services may change the anticipated resource plans and even be redeployed in lower priced data centers. Therefore, the design principle of the invention is to respectively optimize the distribution of the geographic distribution interactive service workload and adjust the corresponding resource plan according to the resource requirements.
The invention discloses a geographical distribution interactive service cloud resource collaborative optimization method which comprises the following steps:
step 1: determining the available data centers for each region. Assume that there are R data center regions that provide resources for the operation of the geographically distributed interactive services workload. In each region, there are P cloud service providers and each cloud service provider has deployed a DCiA data center, wherein i ∈ P. order JrRepresenting the number of data centers in region r, we can calculate all available data centers in region r using the following formula (1):
Figure GDA0002542856510000061
step 2: determining that the interactive service can schedule the distribution of all possible data center combinations. Assume that there are m geographically distributed interactive services, each containing more than | R | workloads that need to be distributed to a data center group. For example, a user sends a service request to n (n ≧ R |) data centers, which are referred to as a target data center group. In particular, the n data centers are provided by different regions, i.e., each region needs to provide at least one data center to complete the operation of the workload. Thus, let G denote all possible data center groups provided by the R zones. We use combinatorial mathematical theory to calculate all available data center groups as follows (2):
Figure GDA0002542856510000062
and step 3: designing and appointing a data center resource providing mode. Because the single use of reserved resources or the single use of on-demand resources requires a high resource cost, both reserved resources and on-demand resources are considered. There are three resource usage scenarios based on this premise: reservation only, on-demand and hybrid only. The step 3 specifically comprises the following steps:
step 3.1: using reserved resource pattern only; the data center provides reserved resources, and the cloud consumer firstly makes a reserved contract and pays a prepaid fee according to the service deployment cycle and the compromised resource requirements; after the interactive service is on line, firstly, the reserved resources are distributed according to the actual requirements of the service resources, and if the reserved resources can meet the requirements of the service quality in the deployment period, the service quality requirements of the interactive service can be met only by reserving the resources.
Step 3.2: using only on-demand resource patterns; the actual demand of the service, which is greater than the lease maximum available resource amount due to the change of the request rate, results in the service quality not reaching the requirement, and the additional on-demand resources can be leased on the basis of the reserved resources to reach the requirement of the service quality, which is called a hybrid phase (reserved and on-demand).
Step 3.3: a hybrid mode. Cloud consumers prefer short term deployment of on-demand resources, which only are provided by data centers.
Determining a probability distribution model of price demand corresponding to a stochastic programming algorithm, wherein the price of data center resources is dynamically changed, and the resource demand of geographic distribution interactive services is dynamically changed according to a service request rate, the uncertainty of the price and the demand can be processed by utilizing the stochastic programming, the stochastic programming adopts a group of uncertain parameters described by probability distribution to decompose a problem into a main problem and a series of sub-problems related to probability, solving the main problem can obtain a primary result, then solving the sub-problems one by one to compensate the error of the primary solution of the main problem, and the step 4 comprises the following steps of determining the probability distribution of each sub-problem of the stochastic programming, providing a plan according to the three resources, analyzing the probability attributes of the price and the demand, and leading Λ to be used for solving the sub-problems one by one to compensate the error of thetRepresenting a set of price and demand cases in each RPP (resource plan) time T, Λ is defined as the following formula (3) for all price and demand cases in one WDD (workload schedule) time T:
Figure GDA0002542856510000071
Λ, e.g., set Λ represents a finite number of price and demand cases, denoted Pλ∈[0,1]Show every kind of situationProbability of occurrence of a condition, where λ is a complex variable, λ ═ λ (λ)1,λ2,…,λ|T|)∈Λ。
And 5: and determining a price model according to the resource providing plan formulated in the step 3. The step 5 specifically comprises the following steps:
step 5.1: a reservation contract prepaid price is determined. Suppose a cloud consumer makes a reserved resource contract k at data center j for workload i, the contract k describing a particular configuration of VM instances, let
Figure GDA0002542856510000079
Indicating the VM instance to resource type
Figure GDA0002542856510000072
Is required by
Figure GDA0002542856510000073
Representing resource types
Figure GDA0002542856510000074
The unit price of (c). Adding the resource type prices required by the VM examples to obtain the prepayment price of the work load i for establishing the reservation contract k in the data center j
Figure GDA0002542856510000075
Step 5.2: a dynamic price for the actual running VM instance is determined. When the VM instance is actually used under the lambda condition, the actual use price of the reserved resources and the resources on demand in the time t can be obtained
Figure GDA0002542856510000076
And
Figure GDA0002542856510000077
because the actually used resources need to be charged according to the use time, wherein t is a charging period and is also an RPP period, as shown in the following formula (4):
Figure GDA0002542856510000078
step 5.3: and determining the charging mode and price of the network resources. WAN network resources are billed based on monthly transmission traffic. WAN network resources will be charged at different unit prices when data transmission exceeds the maximum traffic set in the network pricing policy. Let KnRepresenting the relationship between the amount of data transferred and the price per unit. For example, Amazon EC 2's WAN network price policy is: kn{ "first 1 GB": "$ 0.000/GB", "greater than 1G less than 10 TB": "$ 0.090/GB", "greater than 10T less than 40 TB": "$ 0.085/GB", "greater than 40TB less than 100 TB": "$ 0.070/GB", "greater than 100TB less than 350 TB": "$ 0.050/GB",. Thus, let
Figure GDA0002542856510000081
Representing the unit price of WAN network traffic.
If the geographically distributed interactive service is distributed to a data center group, a sufficient number of VM instances (e.g., 16 core CPU, 64G memory and 500G SSD × 10 per VM instance) must be allocated for the workload.
Figure GDA0002542856510000082
Wherein S denotes the request sources of m geographically distributed interactive services, I denotes the workload set of each interactive service, JgRepresents all data center collections of data center group G (G ∈ G).
Figure GDA0002542856510000083
Representing the number of reserved VM instances that make a reservation contract k at data center j for workload i, a cloud consumer may make multiple reservation contracts at one data center.
Figure GDA0002542856510000084
Representing the amount of network data transmission per charging period of the network resource, where TnTypically expressed as 12 network resource billing periods a year. XijRepresenting a binary variable equal to 1 if workload i is assigned to data center j, and 0 otherwise.
Step 5.5: a dynamic cost of the instance running in real time is determined. As can be seen from equation (5), the resource allocation cost includes three parts: reservation
Figure GDA0002542856510000085
Prepaid cost of the quantity VM instance; cost of WAN network usage; the cost of actually running the reserved VM instance and running the on-demand VM instance. Because the VM is run at startup as needed, the cost of actually running a VM instance is dynamically variable. We use CYRepresenting the dynamic cost. If a reserved VM instance
Figure GDA0002542856510000086
Cannot meet the tail delay requirement of the interactive service: (
Figure GDA0002542856510000087
<
Figure GDA0002542856510000088
<Actual demand), then we need extra
Figure GDA0002542856510000089
The number of VM instances to bring the interactive service to the tail delay requirement is as follows (6):
Figure GDA00025428565100000810
step 6: the tail delay of the interactive service is defined and used as the delay constraint of the optimization model. The step 6 specifically comprises the following steps:
step 6.1: the computational delay and the network transmission delay are determined. Order to
Figure GDA00025428565100000811
And
Figure GDA00025428565100000812
respectively representing data processing delays and WAN network transmission delays. The total delay of the service is calculated using the following equation (7).
Figure GDA00025428565100000813
Step 6.2: determining a delay threshold value of each request and establishing a delay threshold value matrix. Each service request has a corresponding delay requirement, embodied as a delay threshold. We consider each service request independently and we consider a geographically distributed interactive service to meet the end-of-service delay requirement if all service requests meet the respective delay requirement. Thus, let
Figure GDA0002542856510000091
And
Figure GDA0002542856510000092
let L (t) denote the transmit to n (n ═ J), representing the processing delay threshold and WAN network transmission delay threshold, respectivelyg|) delay constraint thresholds for service requests of several data centers, we can get the following delay threshold matrix, as shown in equation (8):
Figure GDA0002542856510000093
step 6.3. Using service level agreement to measure whether the tail delay meets QoS (quality of service) requirements under the resource provision plan determination conditions, since the overhead of predicting the delay of each service request in real time and finally reflecting as resource requirements is very large, especially for a large number of interactive service requests, it is difficult to realize in reality, therefore, we estimate and predict the resource requirements of interactive services using the high percentage tail delay as S L A (service level agreement), and this estimation is based on historical requests.for example, if we set x as the high percentile S L A tail delay constraint, the probability that the service delay does not exceed the threshold value is not less than x%, otherwise the service quality cannot be guaranteed.the high percentile S L A tail delay provides a method for the geographically distributed interactive services to meet the service quality requirements.6.3. specifically, step 6.3.1. using probability statistics to perform probability analysis on historical data to obtain the initial relation between the resource provision plan and S L A tail delay.considering that a request source completes the work load i within t time, the step 6.3.1. the step of performing the probability calculation of the final step of using probability statistics to calculate the following steps:
step 6.3.1.1: and (4) calculating the probability statistical analysis of the delay and the network transmission delay. Order to
Figure GDA0002542856510000094
Representing the probability that the computation delay of workload i at data center j does not exceed the computation delay threshold, order
Figure GDA0002542856510000095
Indicating the probability that the WAN network transmission delay does not exceed the transmission delay threshold. Operation of
Figure GDA0002542856510000096
Represents a method for calculating the processing probability delay of a data center when
Figure GDA0002542856510000097
If so, the result is 1, otherwise it is 0. Similarly, operate
Figure GDA0002542856510000098
For estimating the network probabilistic delay, the following equation (9) is used:
Figure GDA0002542856510000099
the average delay probability values obtained from a large number of requested delay records should be close to the expected value of delay according to the majority rule (LL N). when more requested records are considered in the estimation, the estimation of the delay probability will be more accurate.
Step 6.3.1.2: the probabilistic delay for each request is calculated. The delay of interactive services mainly consists of two parts: a) data center processing delays. b) WAN network transmission delay. The convolution function mathematically produces a third function representing the superposition of the first two functions and is therefore generally considered herein to be the superposition of the computation delay and the network delay. For workload i, the expected probability delay for each service request sent to data center j can be found using a convolution function, as follows (10):
Figure GDA0002542856510000101
where the operator "", represents a convolution method.
Step 6.3.1.3: and calculating probability tail delay. Interactive services require sending requests to a specific data center group to complete the corresponding workload, so the probability tail delay for all service requests of an interactive service s should be averaged across the data center groups as shown in the following equation (11):
Figure GDA0002542856510000102
wherein Fsg(t) probabilistic tail delay, F, for sending an Internet request for interactive service s to data center group gsg(t) as a constrained function of the scheduling decisions and resource provision plans for the interactive services.
And 7: and determining a data center resource scheduling optimization model. Through price modeling and delay constraint modeling, the scheduling of interactive services and resource planning can be obtained by solving the following optimization equation as follows (12) and the like:
Figure GDA0002542856510000103
Figure GDA0002542856510000104
Figure GDA0002542856510000105
Figure GDA0002542856510000111
Figure GDA0002542856510000112
Figure GDA0002542856510000113
Figure GDA0002542856510000114
Figure GDA0002542856510000115
Figure GDA0002542856510000116
wherein (11a) indicates that the workload of each interactive service needs to be allocated to a target data center, (12b) indicates that the tail latency of the interactive service should meet the requirements of the tail latency SA L, (12c) indicates that the allocated resources must be less than or equal to the maximum resource constraint of the data center, (12d) indicates that the number of reserved VM instances actually used does not exceed the maximum constraint of the reservation contract, and (12e-12h) indicates that the number of VM instances is a natural number.
S L A (Service-L event acquisition) Service level Agreement resource allocation plan time (RPP), Workload Distribution Decision time (WDD)
And 7: the overall problem is decomposed into a main problem and a series of sub-problems by using the characteristics of stochastic programming.
The step 7 comprises the following steps:
step 7.1: the deployment cost of the candidate data center for each area is calculated using equation (13). Taking the obtained initial cost as a weight factor of workload scheduling, selecting a data center group with the least cost by utilizing integer programming, scheduling interactive services to an optimized data center group, and residing the interactive services in the data center group in each WDD time interval; meanwhile, at each RPP time, solving the resource plan of the target data center group in the time period by utilizing stochastic programming according to the change of the demand as the following formula (13) and the like:
Figure GDA0002542856510000117
Figure GDA0002542856510000121
Figure GDA0002542856510000122
step 7.2: and determining a main random planning problem and a sub-random planning problem. The main question dispatches the interactive service to the data center group with the lowest cost according to the weight value; a series of sub-questions regarding the probability distribution of price and demand dynamically adjust the resource offering plan according to the dynamic resource demand, as shown in equation (14):
Figure GDA0002542856510000123
Figure GDA0002542856510000124
Figure GDA0002542856510000125
Figure GDA0002542856510000126
Figure GDA0002542856510000127
because of the uncertainty in price, the above algorithm needs to be repeated at each WDD time for workload scheduling and resource planning.
And 8: solving an optimization algorithm, recording the resource demand of each interactive service and corresponding delay information, determining the relation between the resource quantity and the delay and predicting the primary workload scheduling; when the service is on-line, the resource demand and the delay are correspondingly changed, and the changes are used for random planning and finally applied to the dynamic adjustment of the resource plan. Step 8 comprises the following steps:
step 8.1: and (4) workload scheduling. Step 9.1 comprises the following steps:
and 8.1.1, firstly, generating probability distribution of VCPU requirements, memory requirements, disk requirements and network utilization rate according to the request records of the geographic distribution interactive service.
Step 8.1.2, taking the generated probability distribution as configuration data, solving a formula (13) and obtaining the corresponding deployment cost of each schedulable data center;
and 8.1.3, initializing a scheduling decision variable, using the initial cost obtained in the step 9.1.2 as a weight value, using a formula (14) to select the data center with the lowest price as a target data center group, wherein the distribution of the workload needs to meet the constraint of the maximum processing capacity of the data center.
Step 8.2: the workload resource plan is adjusted in real time. Step 8.2 comprises the following steps:
step 8.2.1, through scheduling the workload, we can obtain the target data center groups of all interactive services, but the obtained preliminary resource plan is coarse-grained, and the resource plan needs to be adjusted at any RPP time, so that corresponding resource demand probability distribution is generated at each RPP time;
step 8.2.2, taking the generated probability distribution as configuration data, and solving by using the subproblems defined by the formula (15), so as to obtain a resource plan of each workload in the target data center, wherein the resource plan is fine-grained;
step 8.2.3 our algorithm entails selecting a target data center group for the geographically distributed interactive service at each WDD time, and reconfiguring the resource plan at each RPP time, where each WDD time contains multiple RPP times.
The invention selects a plurality of data centers of three cloud service providers of Google computer Engine, Microsoft Azure and Amazon AWS as experimental configuration data, and selects three interactive services of Web Search, Real-time DataAnalysis and Big Query and related service request records as basic test data. The scheduling of workloads and the resource planning of interactive services were separately experimented.
The workload scheduling process comprises the following steps: firstly, generating probability distribution of VCPU requirements, memory requirements, disk requirements and network utilization rate according to request records of geographic distribution interactive services. The experimental results are shown in fig. 1 to 4; taking the generated probability distribution as configuration data, solving the formula (13) and obtaining the corresponding deployment cost of each schedulable data center; and initializing scheduling decision variables, using the obtained initial cost as a weighted value, using the data center with the lowest price as a target data center group, and meeting the constraint of the maximum processing capacity of the data center according to the distribution requirement of the workload.
The flow of the workload resource plan is as follows: through the dispatching of the workload, target data center groups of all interactive services can be obtained, but the obtained preliminary resource plan is coarse-grained, and the resource plan needs to be adjusted at any RPP time, so that corresponding resource demand probability distribution is generated at each RPP time; solving by using the generated probability distribution as configuration data and utilizing the subproblems defined by the formula (15), so as to obtain a resource plan of each workload in the target data center, wherein the resource plan is fine-grained; a target data center group needs to be selected for the geographically distributed interactive service at each WDD time and the resource plan is reconfigured at each RPP time, where each WDD time contains multiple RPP times.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes and modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention.

Claims (10)

1. A geographical distribution interactive service cloud resource collaborative optimization method is characterized by comprising the following steps:
step 1: determining available data centers for each area;
step 2: determining all available data center combinations that the interactive service can mobilize;
and step 3: designing and appointing a data center resource providing mode;
and 4, step 4: determining a price demand probability distribution model corresponding to a stochastic programming algorithm;
and 5: determining a price model according to the resource providing mode plan formulated in the step 3;
step 6: defining tail delay of the interactive service and using the tail delay as a delay constraint of an optimization model;
and 7: determining a data center resource scheduling optimization model;
decomposing the overall problem into a main problem and a series of sub-problems by using the characteristics of random programming;
and 8: solving an optimization algorithm, recording the resource demand of each interactive service and corresponding delay information, determining the relation between the resource quantity and the delay and predicting the primary workload scheduling; when the service is on-line, the resource demand and the delay are correspondingly changed, and the changes are used for random planning and finally applied to the dynamic adjustment of the resource plan.
2. The method for collaborative optimization of cloud resources for geographic distribution interactive services according to claim 1, wherein the step 3 specifically includes the following steps:
step 3.1: using reserved resource pattern only; the data center provides reserved resources, and the cloud consumer firstly makes a reserved contract and pays a prepaid fee according to the service deployment cycle and the compromised resource requirements; after the interactive service is on line, firstly, the reserved resources are distributed according to the actual requirements of the service resources, and if the reserved resources can meet the requirements of the service quality in the deployment period, the service quality requirements of the interactive service can be met only by reserving the resources;
step 3.2: using only on-demand resource patterns; the actual demand of the service is greater than the maximum available resource number of the lease due to the change of the request rate, so that the required service quality can not be achieved, and extra resources on demand are leased on the basis of reserved resources to achieve the requirement of the service quality, wherein the situation is called a hybrid phase;
step 3.3: in the mixed mode, the cloud consumers perform short-term deployment of services according to the resources on demand, and the data center only provides the resources on demand.
3. The method for collaborative optimization of cloud resources for geographic distribution interactive services according to claim 1, wherein the step 5 specifically includes the following steps:
step 5.1: determining a pre-paid price of the reserved contract;
step 5.2: determining a dynamic price of an actual running VM instance;
step 5.3: determining a network resource charging mode and price;
step 5.4: overlapping the cost of the computing resources and the cost of the network resources to obtain a total price;
step 5.5: a dynamic cost of the instance running in real time is determined.
4. The method for collaborative optimization of cloud resources for geographic distribution interactive services according to claim 1, wherein the step 6 specifically includes the following steps:
step 6.1: determining a calculation delay and a network transmission delay;
step 6.2: determining a delay threshold value of each request, and establishing a delay threshold value matrix;
step 6.3: and (4) utilizing a service level agreement to measure whether the tail delay meets the service quality requirement under the condition of determining the resource providing plan.
5. The method for collaborative optimization of cloud resources for geographic distribution interactive services according to claim 4, wherein the step 6.3 specifically includes the steps of:
and 6.3.1, performing probability analysis on the historical data by using a probability statistical method to obtain a preliminary relation between the resource providing plan and the service level agreement S L A tail delay.
6. The method for collaborative optimization of cloud resources for geographic distribution interactive services according to claim 5, wherein the step 6.3.1 specifically includes the steps of:
step 6.3.1.1: calculating delay and network transmission delay probability statistical analysis;
step 6.3.1.2: calculating a probabilistic delay for each request;
step 6.3.1.3: and calculating probability tail delay.
7. The method for collaborative optimization of cloud resources for geographically distributed interactive services according to claim 1, wherein the step 7 comprises the steps of:
step 7.1: calculating the deployment cost of the candidate data center of each area;
step 7.2: and determining a main random planning problem and a sub-random planning problem.
8. The method for collaborative optimization of cloud resources for geographically distributed interactive services according to claim 1, wherein the step 8 comprises the steps of:
step 8.1: scheduling the workload;
step 8.2: the workload resource plan is adjusted in real time.
9. The method for collaborative optimization of cloud resources for geographically distributed interactive services according to claim 8, wherein the step 8.1 includes the steps of:
step 8.1.1 firstly, generating probability distribution of VCPU (virtual central processing unit) requirements, memory requirements, disk requirements and network utilization rate according to the request record of the geographic distribution interactive service;
step 8.1.2, the generated probability distribution is used as configuration data, and the corresponding deployment cost of each schedulable data center is obtained;
step 8.1.3 initializes the scheduling decision variables.
10. The method for collaborative optimization of cloud resources for geographically distributed interactive services according to claim 8, wherein the step 8.2 includes the steps of:
step 8.2.1, through the dispatching of the workload, the target data center groups of all the interactive services are obtained, but the obtained primary resource plan is coarse-grained, and the resource plan needs to be adjusted in the time of one resource plan RPP, so that corresponding resource demand probability distribution is generated in the time of each resource plan RPP;
step 8.2.2, solving by taking the generated probability distribution as configuration data to obtain a resource plan of each working load in the target data center, wherein the resource plan is fine-grained;
step 8.2.3 selects a target data center group for the geographically distributed interactive service at each workload scheduled WDD time, and reconfigures the resource plan at each resource plan RPP time, where each workload scheduled WDD time includes a plurality of resource plan RPP times.
CN201710639715.5A 2017-07-31 2017-07-31 Geographic distribution interactive service cloud resource collaborative optimization method Active CN107395733B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201710639715.5A CN107395733B (en) 2017-07-31 2017-07-31 Geographic distribution interactive service cloud resource collaborative optimization method
PCT/CN2018/072031 WO2019024445A1 (en) 2017-07-31 2018-01-10 Collaborative optimization method for geographic distribution interactive service cloud resource

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710639715.5A CN107395733B (en) 2017-07-31 2017-07-31 Geographic distribution interactive service cloud resource collaborative optimization method

Publications (2)

Publication Number Publication Date
CN107395733A CN107395733A (en) 2017-11-24
CN107395733B true CN107395733B (en) 2020-08-04

Family

ID=60342409

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710639715.5A Active CN107395733B (en) 2017-07-31 2017-07-31 Geographic distribution interactive service cloud resource collaborative optimization method

Country Status (2)

Country Link
CN (1) CN107395733B (en)
WO (1) WO2019024445A1 (en)

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107395733B (en) * 2017-07-31 2020-08-04 上海交通大学 Geographic distribution interactive service cloud resource collaborative optimization method
CN108848514B (en) * 2018-06-20 2021-08-03 中国联合网络通信集团有限公司 Data communication optimization method and data communication optimizer
CN109325266B (en) * 2018-08-29 2023-11-10 天津大学 Response time distribution prediction method for online cloud service
CN111507554B (en) * 2019-01-31 2022-07-22 北京骑胜科技有限公司 Service resource scheduling method, device, equipment and storage medium
TWI723410B (en) 2019-05-31 2021-04-01 伊雲谷數位科技股份有限公司 Cloud resource management system, cloud resource management method, and non-transitory computer-readable storage medium
CN110888687B (en) * 2019-09-27 2023-03-14 华北水利水电大学 Mobile edge computing task unloading optimal contract design method based on contract design
CN111459670A (en) * 2020-03-30 2020-07-28 中科边缘智慧信息科技（苏州）有限公司 Method for performing cooperative processing at different levels of edge calculation
CN111988388B (en) * 2020-08-13 2023-05-16 北京达佳互联信息技术有限公司 Flow distribution method, device, electronic equipment and storage medium
CN112363829B (en) * 2020-11-03 2024-03-29 武汉理工大学 User resource dynamic allocation method based on elastic scale aggregation
CN112783649B (en) * 2021-01-19 2024-03-01 江苏博云科技股份有限公司 Cloud computing-oriented interaction-aware containerized micro-service resource scheduling method
CN113282413B (en) * 2021-05-20 2024-03-05 南京航空航天大学 QoS demand self-adaptive resource allocation method in vehicle edge computing network
CN113626162A (en) * 2021-07-09 2021-11-09 西安电子科技大学 Data center task hybrid deployment method and system based on dynamic resource sharing
CN113642806A (en) * 2021-08-31 2021-11-12 黄杉 Cloud computing-based optimal design service system and method
CN114697209B (en) * 2022-03-30 2023-12-22 广州穗华能源科技有限公司 Cloud edge collaborative computing resource configuration method and configuration system
CN115189999B (en) * 2022-07-20 2023-08-22 贵州电网有限责任公司 System and method for managing cloud computing services

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103002005A (en) * 2011-09-07 2013-03-27 埃森哲环球服务有限公司 Cloud service monitoring system
CN103530801A (en) * 2013-10-29 2014-01-22 东南大学 Method for optimizing costs of multiple data centers based on dynamic pricing strategy
CN104065663A (en) * 2014-07-01 2014-09-24 复旦大学 Auto-expanding/shrinking cost-optimized content distribution service method based on hybrid cloud scheduling model

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130151688A1 (en) * 2011-12-07 2013-06-13 Alcatel-Lucent Usa Inc. Optimization mechanisms for latency reduction and elasticity improvement in geographically distributed data centers
CN105474654B (en) * 2013-03-15 2019-04-26 奥誓公司 Digital media advertisement system and method for the system
CN107395733B (en) * 2017-07-31 2020-08-04 上海交通大学 Geographic distribution interactive service cloud resource collaborative optimization method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103002005A (en) * 2011-09-07 2013-03-27 埃森哲环球服务有限公司 Cloud service monitoring system
CN103530801A (en) * 2013-10-29 2014-01-22 东南大学 Method for optimizing costs of multiple data centers based on dynamic pricing strategy
CN104065663A (en) * 2014-07-01 2014-09-24 复旦大学 Auto-expanding/shrinking cost-optimized content distribution service method based on hybrid cloud scheduling model

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于任务特性的公有云资源交易机制研究;易小萌;《华中科技大学博士学位论文》;20170501;第2-3章 *
移动云计算中联合资源调度算法研究;刘杰;《重庆邮电大学硕士学位论文》;20160608;第2章 *

Also Published As

Publication number Publication date
WO2019024445A1 (en) 2019-02-07
CN107395733A (en) 2017-11-24

Similar Documents

Publication Publication Date Title
CN107395733B (en) Geographic distribution interactive service cloud resource collaborative optimization method
CN111480145B (en) System and method for scheduling workloads according to a credit-based mechanism
Samaan A novel economic sharing model in a federation of selfish cloud providers
EP3015981B1 (en) Networked resource provisioning system
Liu et al. Minimum-cost cloud storage service across multiple cloud providers
CN103761147B (en) The management method and system of calculated examples in a kind of cloud platform
US10886743B2 (en) Providing energy elasticity services via distributed virtual batteries
Cong et al. A survey of profit optimization techniques for cloud providers
KR20180063371A (en) Burst mode control
US11960255B2 (en) Compute load shaping using virtual capacity and preferential location real time scheduling
Dube et al. Yield management for IT resources on demand: analysis and validation of a new paradigm for managing computing centres
US10873188B2 (en) Dynamic hosting capacity system and method for DER integration and management on distribution utility networks
Depoorter et al. Advance reservation, co-allocation and pricing of network and computational resources in grids
Dibaj et al. Cloud computing energy efficiency and fair pricing mechanisms for smart cities
Yao et al. Optimal overbooking policy for cloud service providers: Profit and service quality
Vanmechelen et al. A simulation framework for studying economic resource management in grids
Li et al. Utility driven dynamic resource allocation using competitive markets in computational grid
Prakash A Critical Review on Federated Cloud Consumer Perspective of Maximum Resource Utilization for Optimal Price Using EM Algorithm
Raghuwanshi et al. The new service brokering policy for cloud computing based on optimization techniques
Sampaio et al. Enhancing reliability of compute environments on amazon EC2 spot instances
Gu et al. Reservation schemes for IaaS cloud broker: a time‐multiplexing way for different rental time
CN111080368A (en) Real-time hierarchical selection method for cloud agent aiming at multi-period reserved instances
Cicchiello Analysis, modeling and implementation of cost models for a multi-cloud Kubernetes context
Sigurleifsson et al. An approach for modeling the operational requirements of FaaS applications for optimal deployment
Lee et al. A market-based resource management and qos support framework for distributed multimedia systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant