![SOA
Login Shipping CheckOut
Search Logout
Buy
[Buy more]
3](https://image.slidesharecdn.com/seamsstatic-120621185207-phpapp02/85/Seams-2012-Reliability-Driven-Dynamic-Binding-via-Feedback-Control-9-320.jpg)
![Adaptation via dynamic
binding
Login Shipping CheckOut
Search Logout
UPS DHL
Buy
[Buy more]
4](https://image.slidesharecdn.com/seamsstatic-120621185207-phpapp02/85/Seams-2012-Reliability-Driven-Dynamic-Binding-via-Feedback-Control-10-320.jpg)
Seams 2012: Reliability-Driven Dynamic Binding via Feedback Control
- 1. Reliability-Driven Dynamic Binding via Feedback Control A. Filieri, C. Ghezzi, A. Leva, M. Maggio
- 3. Motivation Usage profile Running System 2
- 4. Motivation Usage profile Network Running System 2
- 5. Motivation Usage profile Network Running System 3rd parties 2
- 6. Motivation Usage profile Network Running System QoS goals 3rd parties 2
- 7. Motivation Usage profile Network Running System QoS goals 3rd parties 2
- 8. Motivation Usage profile Network Deal with continuous changes Running System QoS goals 3rd parties 2
- 9. SOA Login Shipping CheckOut Search Logout Buy [Buy more] 3
- 10. Adaptation via dynamic binding Login Shipping CheckOut Search Logout UPS DHL Buy [Buy more] 4
- 11. Goal 5
- 12. Goal Make the system continuously provide desired reliability 5
- 13. Goal Make the system continuously provide desired reliability i.e. the probability of successfully accomplishing the assigned task 5
- 14. Goal Make the system continuously provide desired reliability i.e. the probability of successfully accomplishing the assigned task = ¯ 5
- 15. Goal Make the system continuously provide desired reliability i.e. the probability of successfully accomplishing the assigned task = ¯ ¯ 5
- 16. Goal Make the system continuously provide desired reliability i.e. the probability of successfully accomplishing the assigned task = ¯ ¯ max ( ) 5
- 17. State of the art Shipping UPS DHL 6
- 18. State of the art • Heuristics Shipping UPS DHL • Optimization 6
- 19. State of the art • Heuristics Shipping Fast, but no guarantees UPS DHL • Optimization Best decision, but slow 6
- 20. Our proposal Exploit established control theory to get efficient, effective, and scalable dynamic selection 7
- 21. What’s the model w S* 8
- 22. What’s the model S1 w S* S2 8
- 23. What’s the model S1 S w S* S2 F 8
- 24. What’s the model r1 S1 S w 1-r2 S* 1-r1 S2 F r2 8
- 25. What’s the model r1 S1 S p w 1-r2 S* 1-r1 1-p S2 F r2 8
- 26. What’s the model r1(k) S1 S p(k) w(k) 1-r2(k) S* 1-r1(k) 1-p(k) S2 F r2(k) 9
- 27. What’s the model r1(k) S1 S p(k) w(k) 1-r2(k) S* 1-r1(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 28. What’s the model n1(k) r1(k) S1 S n*(k) p(k) w(k) 1-r2(k) S* n2(k) 1-r1(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 29. What’s the model n1(k) , R1 r1(k) S1 S n*(k), R* p(k) w(k) 1-r2(k) S* n2(k) , R2 1-r1(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 30. What’s the model n1(k) , R1 r1(k) S1 S n*(k), R* p(k) w(k) 1-r2(k) S* n2(k) , R2 1-r1(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 31. What’s the model n1(k) , R1 nS(k) r1(k) S1 S n*(k), R* p(k) w(k) 1-r2(k) S* n2(k) , R2 1-r1(k) nF(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 32. What’s the model n1(k) R1 , nS(k) r1(k) S1 S n*(k) R* , p(k) w(k) 1-r2(k) S* n2(k), R2 nF(k) 1-r1(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 33. What’s the model n1(k) , R1 nS(k) r1(k) S1 S n*(k), R* p(k) w(k) 1-r2(k) S* n2(k) , R2 1-r1(k) nF(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 34. What’s the model n1(k) , R1 nS(k) r1(k) S1 S n*(k), R* p(k) w(k) 1-r2(k) S* n2(k) , R2 1-r1(k) nF(k) 1-p(k) S2 F r2(k) Sampling time: Ts 9
- 35. Global picture S1 p w S* 1-p S2 p n1,n2, nS,nF Controller 10
- 36. Controller Reliability of the system: + 11
- 37. Controller Reliability of the system: + Controller’s goal: 11
- 38. Controller Reliability of the system: + Controller’s goal: = ¯ + 11
- 39. Controller Reliability of the system: + Controller’s goal: min( ¯ ) + 11
- 40. Controller Reliability of the system: + Controller’s goal: min( ¯ ) + Controller’s output: 11
- 41. How to design the controller? 12
- 42. How to design the controller? The system has to follow its set point 12
- 43. How to design the controller? The system has to follow its set point The system is not linear 12
- 44. How to design the controller? The system has to follow its set point The system is not linear What are the disturbances of the process? 12
- 45. How to design the controller? The system has to follow its set point The system is not linear What are the disturbances of the process? 12
- 46. Disturbances 13
- 47. Disturbances 1.0 0.8 0.6 0.4 0.2 Fluctuation 0 0 5 10 15 20 25 30 35 Time step 13
- 48. Disturbances 1.0 0.8 0.6 0.4 0.2 Smooth 0 0 5 10 15 20 25 30 35 Time step 1.0 0.8 0.6 0.4 0.2 Fluctuation 0 0 5 10 15 20 25 30 35 Time step 14
- 49. Disturbances 1.0 0.8 0.6 0.4 0.2 Sharp 0 0 5 10 15 20 25 30 35 Time step 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.2 Fluctuation 0.4 0.2 Smooth 0 0 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 Time step Time step 15
- 50. Fluctuation 16
- 51. Fluctuation Equilibrium n = (n, p, ¯) ¯ ¯ ¯ r 16
- 52. Fluctuation Equilibrium n = (n, p, ¯) ¯ ¯ ¯ r Linearizing the system around the equilibrium: 16
- 53. Fluctuation Equilibrium n = (n, p, ¯) ¯ ¯ ¯ r Linearizing the system around the equilibrium: S1 w S2 16
- 54. Fluctuation Equilibrium n = (n, p, ¯) ¯ ¯ ¯ r Linearizing the system around the equilibrium: S1 w S2 Standard PI controller 16
- 55. Smooth and sharp changes Auto-tuner: decide the configuration of the PI to cope with the given equilibrium 17
- 56. Smooth and sharp changes Auto-tuner: decide the configuration of the PI to cope with the given equilibrium Trade-off between responsiveness and overshooting avoidance 17
- 57. Smooth and sharp changes Auto-tuner: decide the configuration of the PI to cope with the given equilibrium Trade-off between responsiveness and overshooting avoidance Limitation: the goal has to be feasible 17
- 58. Multiple alternatives 18
- 59. Multiple alternatives C0 p 1-p 18
- 60. Multiple alternatives C0 p 1-p C0 C0 p 1-p p 1-p 18
- 61. Multiple alternatives C0 p 1-p Level 1 Ts Multirate controller C0 C0 p 1-p p 1-p Level 2 Ts/2 18
- 62. Example C0 p 1-p C0 C0 p 1-p p 1-p 19
- 63. Example C0 p 1-p C0 C0 p 1-p p 1-p .5 .7 .6 .95 19
- 64. Example C0 p 1-p Goal: .9 C0 C0 p 1-p p 1-p .5 .7 .6 .95 19
- 65. Example C0 p 1-p Goal: .9 C0 C0 = . p 1-p p 1-p .5 .7 .6 .95 19
- 66. Example C0 p 1-p Goal: .9 C0 C0 = . = . . p 1-p p 1-p .5 .7 .6 .95 19
- 67. Example C0 = . p 1-p Goal: .9 C0 C0 = . = . . p 1-p p 1-p .5 .7 .6 .95 19
- 68. Validation • Matlab simulation • Java stand-alone • J2EE with Spring and AOP 20
- 69. Validation 21
- 71. Conclusions Effective Efficient Scalable Formally grounded Trade-off reliability/performance Improve AT Best tree balancing Other quantitative properties 22
- 72. Try it @Home http://filieri.dei.polimi.it/publications/2012-seams/ Partially funded by the European Commission, Programme IDEAS-ERC, Project 227977-SMScom 23
- 73. Control Equations { 24
- 74. Control Equations { n( ) = n( ) r( ) +P( ) · r( ) + w( ) r( ) = min{tm , n( )} 24
- 75. Control Equations { n( ) = n( ) r( ) +P( ) · r( ) + w( ) r( ) = min{tm , n( )} ( ) ( ) ( )= ( ) ( )+ ( ) ( ) 24
- 76. Control Equations { n( ) = n( ) r( ) +P( ) · r( ) + w( ) r( ) = min{tm , n( )} ( ) ( ) ( )= ( ) ( )+ ( ) ( ) Set point: ¯ 24
- 77. Fluctuation 25
- 78. Fluctuation Equilibrium n = (n, p, ¯) ¯ ¯ ¯ r 25
- 79. Fluctuation Equilibrium n = (n, p, ¯) ¯ ¯ ¯ r n( ) = A n( )+ Linearized B p( ) + B r( ) system y( ) = C n( ) 25
- 80. Fluctuation Equilibrium n = (n, p, ¯) ¯ ¯ ¯ r n( ) = A n( )+ Linearized B p( ) + B r( ) system y( ) = C n( ) Z-transform ( )= 25
- 81. Controller 26
- 82. Controller Error ( )=¯ ( ) 26
- 83. Controller Error ( )=¯ ( ) Standard PI ( ) = ( )+ ( )· ( ) ( ) = ( )+ · ( ) 26