Th`ese de Doctorat de l'université Paris VI Pierre et Marie Curie Mlle ...

More documents

Recommendations

Info

same time, resource utilization and network revenue. Therefore, in this thesis, we focus on the issue of dynamic bandwidth allocation; we propose and evaluate a new service model that considers users’ requirements. Moreover, we develop a broad set of dynamic bandwidth allocation algorithms that achieve a high network utilization and satisfy users’ preferences. 1.1 Thesis Contribution In this thesis, we first propose a new service model that provides quantitative per-flow bandwidth guarantees, where users subscribe for a guaranteed transmission rate. Moreover, the network periodically individuates unused bandwidth and proposes short-term contracts where extra-bandwidth is allocated and guaranteed exclusively to users who are willing to pay for it to transmit at a rate higher than their subscribed rate. To implement this service model we propose a distributed provisioning architecture composed by core and edge routers; core routers monitor bandwidth availability and periodically report this information to ingress routers using signalling messages like those defined in [1, 2]. Moreover, if persistent congestion is detected, core routers notify imme- diately ingress routers. Ingress routers perform a dynamic tracking of the effective number of active connections, as proposed in [8, 9], as well as of their actual sending rate. Based on such information and that communicated by core routers, ingress routers allocate network resources. For this purpose, we develop a broad set of efficient dynamic bandwidth allocation algorithms that take into account online measured traffic statistics as well as users’ profile and willingness to acquire extra-bandwidth based on their bandwidth utility function. Further, we propose a novel and flexible mathematical model for the extra-bandwidth allocation problem. This model assumes the exact knowledge of the future traffic offered to the network; it allows to maximize the total network revue, and provides theoretical 18
upper bounds to the performance achievable by any online dynamic bandwidth allocation algorithm. We evaluate by simulation the performance of our proposed bandwidth allocation algorithms in realistic network scenarios. Numerical results show that our algorithms and service model allow to achieve better performance than statically provisioned networks both in terms of total accepted load and network revenue. Furthermore, simulation results show that the proposed algorithms approach, in several network scenarios, the ideal performance provided by the mathematical model. In summary, this thesis makes the following main contributions: • The definition of a new service model that takes into account users’ profile and traffic statistics. • The proposition of a broad set of efficient heuristic algorithms for dynamic bandwidth allocation that aim at maximizing both users utility and network revenue. • The illustration of a mathematical model for the bandwidth allocation problem; the solution of this model maximizes the network revenue and allows to obtain an upper bound on the performance achievable by any online allocation algorithm. 1.2 Thesis Structure Thethesisisstructuredasfollows: Chapter 2 discusses the work related to the dynamic bandwidth allocation problem and compares our approach to the literature. Chapter 3 introduces the proposed service model that allows to maximize at the same time users’ utility and total network revenue. To implement such service model, this Chapter further proposes a dynamic provisioning architecture and a set of control messages that assure communication between network elements. 19
Page 1 and 2: Thèse de Doctorat de l’universit
Page 3 and 4: To my family.
Page 5 and 6: Abstract Efficient dynamic resource
Page 7: Acknowledgements I would like to th
Page 10 and 11: 4.2 Connections Classification . .
Page 13 and 14: Résumé delaThèse 1. La probléma
Page 15 and 16: supérieure sur les performances qu
Page 17 and 18: compte les fonctions d’utilité d
Page 19 and 20: (Policy Decision Point), d’un ens
Page 21 and 22: ligne pour individualiser la bande
Page 23 and 24: 6. Nous proposons trois algorithmes
Page 25 and 26: appelons débit de crête. Six sour
Page 27 and 28: 6. Conclusion générale et Perspec
Page 29: Chapter 1 Introduction The Internet
Page 33 and 34: Chapter 2 Dynamic Bandwidth Allocat
Page 35 and 36: Utility Function Types In a multi-a
Page 37 and 38: comes smaller than the application
Page 39 and 40: of which there exist extensive theo
Page 41 and 42: allows us to allocate the bandwidth
Page 43: Shenker in [10] states that the net
Page 46 and 47: implements such service model by pe
Page 48 and 49: considering different constraints l
Page 50 and 51: 4.1 Problem Statement Let us model
Page 52 and 53: y all network users. Using the nota
Page 54 and 55: in the following Section. The struc
Page 56 and 57: Optimum Bandwidth Allocation Algori
Page 58 and 59: Table 5.1: Pseudo-code specificatio
Page 60 and 61: of all greedy connections that are
Page 63 and 64: Chapter 6 Mathematical Model This C
Page 65 and 66: 1 2 3 4 0000000000 1111111111 00000
Page 67 and 68: Chapter 7 Numerical Results The mai
Page 69 and 70: the two core nodes is equal to 6 Mb
Page 71 and 72: Total Accepted Load (Mb/s) 6 5.5 5
Page 73 and 74: Total Accepted Load (Mb/s) 6 5.5 5
Page 75 and 76: plottedinFigure7.9. Figures 7.10 an
Page 77 and 78: Total Accepted Load (Mb/s) 6 5 4 3
Page 79 and 80: S 2 S 3 S 4 1 S 6 S1 S5 2 3 4 5 D 1
Page 81 and 82:
high network loads, where it achiev
Page 83 and 84:
proposed allocation algorithms in a
Page 85 and 86:
Table 7.3: Peak rate, subscribed ra
Page 87 and 88:
Chapter 8 Performance Upper Bound T
Page 89 and 90:
Table 8.1: Average performance gain
Page 91 and 92:
Total Accepted Load (Mb/s) 7 6.5 6
Page 93 and 94:
Total Accepted Load (Mb/s) 6 5 4 3
Page 95 and 96:
Chapter 9 Conclusion 9.1 Discussion
Page 97:
established by creating a session i
Page 100 and 101:
[7] F. P. Kelly, “Charging and ra
Page 102 and 103:
[26] P. Reichl, S. Leinen, and B. S
Page 105 and 106:
Appendix A Algorithm Implementation
Page 107 and 108:
these latters are implemented using
Page 109 and 110:
• boolean State: the connection s
Page 111 and 112:
(CSTtable) is updated by the ingres
Page 113 and 114:
Table A.2: Java code for updating t
Page 115 and 116:
List of Acronyms ACA Autonomous Com
Page 117 and 118:
List of Figures 1 Notre architectur
Page 119 and 120:
8.1 Performance upper bound for all
Page 121:
List of Tables 5.1 Pseudo-code spec
show all

Th`ese de Doctorat de l'université Paris VI Pierre et Marie Curie Mlle ...

Create successful ePaper yourself

Delete template?

Save as template?