1. Changbin Liu, Lei Shi, Bin Liu Department of Computer Science and Technology, Tsinghua University Proceedings of the Fourth European Conference on Universal Multiservice Networks (ECUMN’07) Chen Bin Kuo (20077202) Young J. Won (20063292)

2. DPNM Lab.  Introduction  NGN traffic classifications and their utility functions  Network utility maximization (NUM)  Numeric results and analysis  Discussion  Conclusion 10/3/20152

3. DPNM Lab.  Next generation network (NGN) must natively support triple-plays.  How to schedule traffic and allocate bandwidth at both backbone and access links.  Designing a scheduling (bandwidth allocation) algorithm is exactly the issue this paper tries to settle. 10/3/20153

4. DPNM Lab.  In industry designing NGN [13][14], the strict- priority scheduling is mostly adopted.  Rigidly favors the voice and video traffic without flexibility  Utility-based solutions  Shenker [1] discussed traffic classifications in IP network from the viewpoint of user utility  Kelly et al. [5][6] applying utility-based methods to scheduling and bandwidth allocation in the objective of Network Utility Maximization (NUM) 10/3/20154

5. DPNM Lab.  No single work has emphasized on the practical issue of scheduling triple-play services under the background of NGN.  Translating this issue into a nonlinear maximization problem with inequality constraints. 10/3/20155

6. a) VoIP traffic b) IPTV traffic c) TCP elastic traffic d) HTTP traffic e) Other UDP traffic 10/3/20156

7. DPNM Lab.  Due to remarkable distinction of QoS requirements in NGN  Classifying NGN traffic into five categories  User utility function is introduced  To measure network performance and user satisfaction degree  Determined by the QoS metrics received in the user end  Including packet delay, jitter and loss rate 10/3/20157

8. DPNM Lab.  Sensitive to packet delay and loss caused by bandwidth insufficiency  Utility function falls into the category of hard real-time kind [1][2][10], with a minimal bandwidth requirement of Bmin1 10/3/20158

9. DPNM Lab.  Utility function is similar to VoIP’s but with some differences  Tolerate occasional delay-bound violations and packet drops  Minimal encoding rate, denoted as Bmin2 is independent of network congestion  Logistic model is used Logistic model 10/3/20159

10. DPNM Lab.  Generated by delay-tolerant TCP applications  Such as file transfer and email  Utility function have been studied by Kelly et al. [6] and other researchers [11][12] 10/3/201510

11. DPNM Lab.  TCP traffic which concerns packet delay  Mainly contains the HTTP traffic generated by web services  Utility function is different from TCP elastic traffic, has a minimum tolerable bandwidth Bmin4 10/3/201511

12. DPNM Lab.  DNS packets, other streaming media traffic, and on- line gaming traffic [17][18]  Delay-sensitive  Every application type has a utility function  The shape of utility function resembles IPTV traffic 10/3/201512

13. DPNM Lab. 10/3/201513 VoIPIPTVTCP elasticHTTPUDP B min 64Kbps100 Kbps24Kbps B max 10Mbps 500Kbps ɛ0.001

14. a) KKT method b) Lagrange multipliers method without KKT conditions 10/3/201514

15. DPNM Lab.  Based on NGN traffic’s utility functions, we can solve the congestion-phased bandwidth allocation issue while conforming to NUM.  Total utility gained on the link is:  Bandwidth allocation is restricted by: 10/3/201515 N : the number of NGN users utilizing this link p i : traffics classes C : the bandwidth of a link (set to 10Gbps) N : the number of NGN users utilizing this link p i : traffics classes C : the bandwidth of a link (set to 10Gbps)

16. DPNM Lab.  Lagrange Multiplier method with KKT (Karush- Kuhn-Tucker) conditionsKKT (Karush- Kuhn-Tucker) conditions  Solving the nonlinear optimization problem  Accurate and comprehensive solution requires substantial complicated computations  Applying simplified form which is enough to ravel NUM problem for triple-plays 10/3/201516

17. DPNM Lab.  Observing NGN traffic’s utility functions  VoIP/IPTV/other UDP traffic’s utility functions are relatively smoother in some points  It is not cost-effective to allocate bandwidth to VoIP/IPTV/other UDP traffic without booming the utility  Turning point (TP) 10/3/201517 Bandwidth IPTV HTTP

18. DPNM Lab.  After finding the TP, we can apply the Lagrange Multipliers method without KKT conditions to solve the NUM problem in (10)Lagrange Multipliers method  Subject to: 10/3/201518

19. a) Data-dominated network b) IPTV-dominated network 10/3/201519

20. DPNM Lab.  Two network scenarios  Current Internet, where HTTP and TCP elastic traffic still dominate the volume  Prospective NGN, where the emerging services, especially the IPTV traffic, will dominate the network  For each scenario, calculate in two situations  Maximal Utility Equalization (MUE)  Maximal Utility In-equalization (MUI) 10/3/201520 V1 (VoIP)V2 (IPTV)V3 (TCP elastic)V4 (HTTP)V5 (other UDP) MUE11111 MUI1911.52

21. DPNM Lab.  Data-dominated network  According to recent trace observation [15]  IPTV-dominated network 10/3/201521 VoIPIPTVTCP elasticHTTPother UDP Traffic proportions 10% 50%20% VoIPIPTVTCP elasticHTTPother UDP Traffic proportions 10%50%10%20%10%

22. DPNM Lab. 10/3/201522

23. DPNM Lab. 10/3/201523

24. DPNM Lab.  Previous bandwidth allocation schemes for triple- play services mostly adopt the strict-priority scheduling  Highest priority to VoIP traffic  Second highest priority to IPTV and lowest priority to others  In this paper  Highest priority to VoIP traffic  Assigning IPTV traffic with second-highest priority is not well supported from the objective of NUM  Suggesting that ISP charges more about IPTV services (future work) 10/3/201524

25. DPNM Lab.  Studied the problem of scheduling and bandwidth allocation for triple-play services in the objective of NUM.  Presenting theoretical method to compute bandwidth allocation results  Results:  VoIP and other low-throughput UDP traffic can always be guaranteed of sufficient bandwidth  As congestion becomes severer, IPTV’s bandwidth decreasing quickly  TCP elastic and HTTP traffic experience exponential bandwidth degradations when congestion degree increases 10/3/201525

26. 10/3/201526