QoS-driven Web Services Selection in Autonomic Grid Environments. Danilo Ardagna Gabriele Giunta Nunzio Ingraffia Raffaela Mirandola Barbara Pernici

Size: px

Start display at page:

Download "QoS-driven Web Services Selection in Autonomic Grid Environments. Danilo Ardagna Gabriele Giunta Nunzio Ingraffia Raffaela Mirandola Barbara Pernici"

Dominick Lynch
8 years ago
Views:

1 QoS-driven Web Services Selection in Autonomic Grid Environments Danilo Ardagna Gabriele Giunta Nunzio Ingraffia Raffaela Mirandola Barbara Pernici

2 Introduction In SOA, complex applications can be composed as business processes invoking a variety of available WS Develop applications by specifying component WS only through their required functional characteristics and non-functional constraints Select WS during process execution from a registry of available services, guaranteeing QoS constraints D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 2

characteristics and non-functional constraints Select WS during process execution from a registry of available

3 Satisfying QoS constraints QoS requirements are difficult to satisfy especially for the high variability of workloads and need autonomic computing self-managing techniques Grid provides basic mechanisms to manage a service center infrastructure, simplifying the reconfiguration of the physical resources We propose a reference framework to support the execution of e-business applications in autonomic grid environments The problem of selection of WS will be analyzed in depth, assuming variable QoS profiles for WS and the long term execution of the composed services D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 3

reference framework to support the execution of e-business applications in autonomic grid environments The problem of selection of WS will be analyzed in depth,

4 Outline An e-business case study Preliminary definitions Reference Framework Optimization Problem Formulation Experimental Analyses Conclusions and Future work D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 4

Experimental Analyses Conclusions and Future work D.

5 Case study VO composed by a set of small and medium enterprises ERP software execution (AgileERP) The cost allocation process 1,500 Cn Cn = Transfer Cost Center Cn.1, Cn.2, Cn.3, = Riceiver Cost Center Level n Level n+1 Cn.1 Cn.2 Cn ,050 Cn.2.1 Cn.3.1 D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 5

3, = Riceiver Cost Center Level n 0.1 0.2 0.7 Level n+1 Cn.1 Cn.2 Cn.3 150 300 1,050 Cn.2.1 Cn.3.1 D.

6 Case study VO composed by a set of small and medium enterprises ERP software execution (AgileERP) The cost allocation process 1,500 Cn Cn = Transfer Cost Center Cn.1, Cn.2, Cn.3, = Riceiver Cost Center Level n Level n+1 Cn.1 Cn.2 Cn , Cn.2.1 Cn.3.1 D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 6

3, = Riceiver Cost Center Level n Level n+1 Cn.1 Cn.2 Cn.3 150 300 1,050 1 0.1 Cn.2.1 Cn.3.1 D.

7 Case study VO composed by a set of small and medium enterprises ERP software execution (AgileERP) The cost allocation process 1,500 Cn Cn = Transfer Cost Center Cn.1, Cn.2, Cn.3, = Riceiver Cost Center Level n Level n Cn.1 Cn.2 Cn Cn.2.1 Cn D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 7

3, = Riceiver Cost Center Level n Level n+1 150 Cn.1 Cn.2 Cn.3 945 Cn.2.1 Cn.3.1 300 105 D.

8 The cost allocation process CostHierarchy. BalanceVerification(CC[]) Probability=0.95 OK Check Probability=0.05 Error CostAccounting. AllocateCosts CostHierarchy. NotificationError Report.GenerationPDF Report.GenerationExcel CostAccounting. ConsolidateCosts D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 8

NotificationError Report.GenerationPDF Report.GenerationExcel CostAccounting.

9 The allocate costs sub-process For i=1.. CC FALSE i< CC TRUE CostAccounting. CalculateCharge Balance(CC[i]) FALSE CostAccounting. CalculateTotalCost Transferred (AR) AR=0 TRUE CostAccounting. AllocateCosts CostAccounting. NavigateCost Hierarchy End of CH FALSE TRUE D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 9

CalculateTotalCost Transferred (AR) AR=0 TRUE CostAccounting.

10 Preliminary Definitions A Web service is modeled as a software component which implements a set of operations Service Registries stores also for each operation o of a Web service j, the values of QoS q n j,o(t) and the number of instances j,o (t) that can be executed during a specific time inistant A composite service is specified as a high-level business process in BPEL language in which the composed Web service is specified at an abstract level Abstract Web service (task t i ) and concrete Web services (ws j ) D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 10

composite service is specified as a high-level business process in BPEL language in which the composed Web service is specified at an abstract level

11 Preliminary Definitions BPEL annotations: global and local constraints on quality dimensions cycles maximum number of iterations and expected frequency of execution of conditional branches user preferences, {ω 1, ω 2,..., ω N },Σω n = 1 Web service dependency constraints Constraints and BPEL annotations specified by WS-Policy: Global Contraints: ExeTime 2 days ExeTime(BalanceVerification)+ExeTime(NotificationError) 20 sec Web service dependency: BalanceVerification, NotificationError D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 11

.., ω N },Σω n = 1 Web service dependency constraints Constraints and BPEL annotations specified by WS-Policy: Global Contraints: ExeTime 2 days

12 QoS Profiles q n j,o(t) u=13 u=28 Δ T 2T t D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 12

13 Grid Reference Framework D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 13

14 Resource Management in Grid Environments Resource management introduces two different optimization problems which corresponds to the VOs (providers) and users perspectives: each VO would like to maximize the SLA revenues and the use of physical resources the end user is interested in the maximization of the QoS of the composed service execution D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 14

revenues and the use of physical resources the end user is interested in the maximization of the QoS of the

15 Local Resource Allocation Each VO needs to allocate local grid physical resources to different Web service operation invocations ws j,o in order to maximize the revenues from SLA, while minimizing resource management costs Autonomic techniques allow the dynamic allocation of physical resources among different Web services on the basis of short-term demand estimates The local resource allocation is performed periodically with period Δ (e.g., minutes) D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 15

of physical resources among different Web services on the basis of short-term demand estimates The local resource allocation is performed

16 Local Resource Allocation Short-term workload predictor ^ Ν j,o (t) Workload statistics Local Resource Allocator Grid Monitor Ν j,o (t) q n j,o (t) f j,o (t) Long-term workload predictor Historical workload statistics Local Resource Allocator Algorithm (Simulation) Ν j,o (t) q n j,o (t) T Δ T D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 16

Historical workload statistics Local Resource Allocator Algorithm (Simulation) Ν j,o (t) q n j,o

17 Maximizing QoS for the End User Execution Path (ep k ): a set of tasks {t 1,t 2,,t I }, such that t 1 is the initial task, t I is the final task and no t i1, t i2 belong to alternative branches Execution Plan: a set of ordered triples {(t i, ws j,o,x i )}, indicating that task t i included in ep k is executed at time instant x i by invoking ws j,o Global Plan: a set of ordered triples {(t i, ws j,o,x i )}, which associates every task t i to a given WS invocation ws j,o at time instant x i q n (k): n-quality dimension value along execution path k D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 17

instant x i by invoking ws j,o Global Plan: a set of ordered triples {(t i, ws j,o,x i )}, which associates every task t i to a given WS invocation ws j,o at

18 Execution Paths CostHierarchy. BalanceVerification(CC[]) Execution path ep 1 Execution path ep 2 CostAccounting. CalculateCharge Balance(CC[1]) CostAccounting. CalculateTotalCost Transferred (AR) CostHierarchy. BalanceVerification(CC[]) CostAccounting. AllocateCosts CostHierarchy. NotificationError CostAccounting. NavigateCost Hierarchy Report.GenerationPDF Report.GenerationExcel CostAccounting. ConsolidateCosts D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 18

BalanceVerification(CC[]) CostAccounting. AllocateCosts CostHierarchy. NotificationError CostAccounting.

19 Problem Formulation The WSC problem is multi-objective and a simple additive weighting technique is used to evaluate the overall value of QoS from multiple quality dimensions Decision variables: y i,j,o,u equals 1 if the Web service operation ws j,o executes task t i during the time interval u, 0 otherwise w i,u is equal to 1 if the task t i is executed during time interval u, 0 otherwise x i start time of task t i D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 19

j,o executes task t i during the time interval u, 0 otherwise w i,u is equal to 1 if the task t i is executed during time

20 Maximizing QoS for the End User D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 20

21 Maximizing QoS for the End User The problem is NP-hard, equivalent to a Multiple choice Multiple Dimension Knapsack Problem D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 21

22 Experimental Results Analyses performed on a wide set of randomly generated problem instances Problem solution obtained by CPLEX, a state of the art integer linear programming solver (execution time limited to one minute, approximate solution) Percentage gap between the approximate and global optimum solutions D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 22

23 Conclusions and Future work This paper presents a framework for the development of e-business applications built on autonomic grid computing infrastructure Service selection and composition is performed to guarantee that the overall quality perceived by the user is maximized Implementation of the proposed approach and its validation on a real industrial test bed Optimization of a set of instances D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 23

24 Thanks! Any questions? D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 24

25 Local Constraints D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 25

26 Web Service Dependency Constraints D. Ardagna, G. Giunta,, N. Ingraffia, R. Mirandola,, B. Pernici - GADA rd November 26

Run-time Resource Management in SOA Virtualized Environments. Danilo Ardagna, Raffaela Mirandola, Marco Trubian, Li Zhang

Run-time Resource Management in SOA Virtualized Environments Danilo Ardagna, Raffaela Mirandola, Marco Trubian, Li Zhang Amsterdam, August 25 2009 SOI Run-time Management 2 SOI=SOA + virtualization Goal: