Utility-based Optimal Service Selection for Business Processes in Service Oriented Architectures

Transcription

1 Utility-based Optimal Service Selection for Business Processes in Service Oriented Architectures Vinod K. Dubey Daniel A. Menascé Web Services (ICWS), 2010 IEEE International Conference on. IEEE, 2010 Summarized by: Noor Bajunaid 1

2 Introduction Service Oriented Architectures allow service providers to provide similar functionalities with different QoS and cost. There is a need for server provider selection algorithms that optimize a utility function under constraints, efficiently. 2

3 Problem definition A business process B, with N activities, a i,,a n subject to Maximize U(E[R(z)],A(z),X(z)) E[R(z)] R max A min A(z) 1 X(z) X min C(z) C max z Z 3

4 BPEL <sequence> <invoke a1> <switch> <case q1> <flow> <invoke a2> <sequence> <invoke a3> <invoke a4> </sequence> </flow> <case q2=(1-q1)> <invoke a5> </switch> <invoke a6> </sequence> 1: a1 0: sequence 2: switch 9: a6 q1 q2 3: flow 8: a5 4: a2 5: sequence 6: a3 7: a4 4

5 Utility Functions 5

6 Computation of End-to-End QoS Metrics 6

7 Availability 0: sequence 1: a1 2: switch 9: a6 4: a2 q1 3: flow q2 8: a5 5: sequence 6: a3 7: a4 A = A1 * {q1* A2 * [A3 * A4] + q2 * A5} * A6 7

8 Computation of End-to-End QoS Metrics 8

9 Throughput 0: sequence 1: a1 2: switch 9: a6 4: a2 q1 3: flow q2 8: a5 5: sequence 6: a3 7: a4 X = min{x1, (q1* min{x2, X3, X4}), q2 * X5}, X6} 9

10 Computation of End-to-End QoS Metrics **Menascé, Daniel A., Emiliano Casalicchio, and Vinod Dubey. "On optimal service selection in service oriented architectures." Performance Evaluation 67.8 (2010):

11 Execution Time 0: sequence 1: a1 2: switch 9: a6 4: a2 q1 3: flow q2 8: a5 5: sequence 6: a3 7: a4 R = R1 + q1 * max {R2, (R3 + R4)} + q2 * R5 + R6 11

12 Computation of End-to-End QoS Metrics **Menascé, Daniel A., Emiliano Casalicchio, and Vinod Dubey. "On optimal service selection in service oriented architectures." Performance Evaluation 67.8 (2010):

13 Cost 0: sequence 1: a1 2: switch 9: a6 4: a2 q1 3: flow q2 8: a5 5: sequence 6: a3 7: a4 C = C1 + q1 * (C2 + C3 + C4) + q2 * C5 + C6 13

14 Optimal Service Selection 1) Extended JOSeS Algorithm: optimal solution efficient for moderate complicity 2) HCB Heuristic Algorithm near-optimal solution efficient even for large set of services 14

15 Extended JOSeS Algorithm Extends Jensen-based Optimal Service Selection Jensen s inequality: E[max{R 1,, R n }] max{e[r 1 ],, E[R n ]} It is expensive to compute E[max{R 1,, R n }]. Jensens s inequality provides a lower bound that is easier to compute. If the lower bound exceeds the maximum execution time, we ignore the allocation and avoid the expensive computation. 15

16 Extended JOSeS Algorithm If sub-allocation (s1,, sk), k N, violates a constraint, it can be discarded without the need for selecting SPs for activities of order > k. 16

17 Extended JOSeS Algorithm 17 Let lk be the list of SPs for ak: next(k) returns the next, not yet evaluated, SP in lk, or returns null if all the SPs in lk were already evaluated. reset(k) sets all SPs in all lists lj (j = k,..., N ) as not-visited. **Menascé, Daniel A., Emiliano Casalicchio, and Vinod Dubey. "On optimal service selection in service oriented architectures." Performance Evaluation 67.8 (2010): **

18 Extended JOSeS Algorithm S1,1 S1,2 S1,3 S1,1 S1,2 S1,3 S1,1 S1,2 S1,3 S2,1 S2,2 S2,1 S2,2 S2,1 S2,2 S3,1 S3,2 S3,3 S3,1 S3,2 S3,3 S3,1 S3,2 S3,3 S1,1 S1,1 S2,1 violation 18

19 Extended JOSeS Algorithm S1,1 S1,2 S1,3 S1,1 S1,2 S1,3 S1,1 S1,2 S1,3 S2,1 S2,2 S2,1 S2,2 S2,1 S2,2 S3,1 S3,2 S3,3 S3,1 S3,2 S3,3 S3,1 S3,2 S3,3 S1,1 S2,2 S1,1 S2,2 S3,1 S1,1 S2,2 S3,2 19

20 Extended JOSeS Algorithm S1,1 S1,2 S1,3 S1,1 S1,2 S1,3 S2,1 S2,2 S2,1 S2,2 S3,1 S3,2 S3,3 S3,1 S3,2 S3,3 S1,1 S2,2 S3,3 S1,2 Allocations that violate constraint will reduce the number of examined points 20

21 HCB Heuristic Algorithm Hill-climbing based: Define a neighborhood of an allocation Move to the best allocation in the neighborhood Repeat until near-optimum solution is found or maximum number of starts 21

22 HCB Heuristic Algorithm Neighborhood: for each activity, replace the SP with the other SPs that will maximize improvement in each QoS metric 22

23 HCB Heuristic Algorithm 23

24 HCB Heuristic Algorithm 24

25 Experimental Evaluation 1. Determine how effective is the heuristic solution compared to the optimal. 2. Compare the number of points examined by each algorithm 3. Compare both algorithms over a wide range of parameters 25

26 Experimental Evaluation 50 BPEL business processes. 6-9 activities with different construct (sequence, flow, switch) 2-7 SPs per activity. Constraints strength varied from 10% to 40% each combination was ran through JeSOS once,and through HCB 30 times 26

27 Experimental Evaluation QoS metrics of each SP for each activity are given: (E[R],A, X) CTotal = C(r) + C(X) + C(A) 27

28 Experimental Evaluation stricter constraints reduce the size of the neighborhood and decrease the breadth of the search. 28

29 Experimental Evaluation As CS increases, more sub-allocations are prematurely declared unfeasible. JeSOS will examine significantly less points and take less time. 29

30 Experimental Evaluation For a complex Business process and 7 SPs, HCB achieved 99.97% of optimal utility by examining 100 points. JeSOS examined more than 10,000,000 points! 30

31 Experimental Evaluation HCB scalability for large number of SPs/activity (50-400). Regression shows that the number of examined points increases linearly with umber of SPs. 31

32 Conclusion The most important conditions for JeSOS to be efficient are: simple business process structure. limited number of server providers stronger constraints. HCB is potentially very efficient for autonomic nearoptimal resource allocation. 32