A Case for Relative Differentiated Services and the Proportional Differentiation Model

1. A Case for Relative Differentiated Servicesand the Proportional Differentiation Model Authors: Constantinos Dovrolis and Parameswaran Ramanathan University of Wisconsin-Madison Source:IEEE Network • September/October 1999 Reporter: 吳佳幸 湯美蓮 黃瑋瀅

2. Evolution • QoS 概述 • Best- effect • Intserv • Diffserv absolute service differentiation relative service differentiation

3. Outline • The Integrated Services Approach • The Differentiated Services Approach • Relative Differentiated Services Models • Two Features for Relative Service Differentiation • The Proportional Differentiation Model • Comparison to Two Other DiffServ Models

4. Outline • Forwarding Mechanisms for Proportional Differentiation • A Scheduler for Proportional Delay Differentiation • A Dropper for Proportional Loss Rate Differentiation • Conclusion

5. The Integrated Services Approach • The integrated services(IntServ) approach focuses on individual packet flows. • The three major components of the IntServ architecture : • admission control unit • the packet forwarding mechanisms • ResourceReservation Protocol(RSVP)

7. The Differentiated Services Approach • The differentiated services(DiffServ) approach is focusing on traffic aggregates. • Two different directions of research on DiffServ • absolute service differentiation • relative service differentiation

9. Differentiated Services vs. the Fat-Dump-Pipe Model • The introduction of Fat-Dump-Pipe Model • inefficient in terms of network economics and resource management • all traffic receives the same, normally very high, quality of service • The comparative of Differentiated Services and Fat-Dump-Pipe Model

11. Three Relative Differentiation Models • Strict prioritization • Starvation • Not controllable • Price Differentiation • Paris Metro Pricing (PMP)

12. Three Relative Differentiation Models (cont.) • Capacity Differentiation • Weighted Fair Queuing (WFQ) where i,j are class ; wi is class i’ weight; λi is class i’ arrival rate;

13. Three Relative Differentiation Models (cont.)

14. Two Features for Relative Service Differentiation • Controllability • meaning that the network operators should be able to adjust the quality spacing between classes based on their pricing or policy criteria • Predictability • in the sense that the class differentiation should be consistent even in short timescales, independent of the variations of the class loads

15. The Proportional Differentiation Model • Performance measure • Average queuing delay • Packet loss rate

16. Comparison to Two Other DiffServ Models • Premium ( or Virtual leased Line) Service : • premium service user is given the guarantee for a nominal bandwidth with minimal queuing delays and losses along a certain network path, independent of the behavior of the rest of the traffic in that path • Assured Service: • it also provides users with bandwidth assurances along certain network paths or in an entire network, but without strict guarantees that this bandwidth will always be available

17. Forwarding Mechanisms for Proportional Differentiation

18. A Scheduler for Proportional Delay Differentiation • waiting time priority (WTP) scheduler • the priority of a packet in queue i at time t is where wi(t) is the waiting time of the packet at time t. The DDPs {δi} determine the rate at which the priority of the packets of a certain class increases with time.

19. A Scheduler for Proportional Delay Differentiation (cont.)

20. A Scheduler for Proportional Delay Differentiation (cont.)

21. A Dropper for Proportional Loss Rate Differentiation • loss history buffer (LHB) • the minimum normalized loss rate • Note that in order to achieve loss rate differentiation in short timescales, we would prefer lower values of K

22. A Dropper for Proportional Loss Rate Differentiation (cont.)

23. Conclusion • The DiffServ architecture are scalable and easy-to-deploy service differentiation mechanisms • Relative differentiation approach can provide different applications and best matches their quality-cost • The proportional differentiation model allows the network operator to control the quality spacing between classes independent of class loads, and can provide consistent class differentiation in short timescales