1. INTERNET QOS: A BIG PICTURE XIPENG XIAO AND LIONEL M. NI, MICHIGAN STATE UNIVERSITY Jinyoung You CS540, Network Architect

2. Motivation  Today’s Internet  Best-effort service  No guarantee as to timeliness or actual delivery  No guarantee of low-delay and low-jitter services  No service classes  No abundant and cheap bandwidth 2

3. Motivation  What’s the QoS?  Guarantee a certain level of performance e.g.) Packet dropping probability, Delay, Jitter, Out-of- order delivery, Error, Congestion  Why QoS?  The network capacity is insufficient Real-time streaming multimedia applications e.g.) VoIP, IPTV Requires fixed bit rate and are delay sensitive 3

4. Motivation  Misunderstanding of QoS  Cannot provide nonexistent bandwidth  Cannot make the network faster  Cannot cure poorly performing network  Contribution of QoS  Provide relative prioritization of traffic 4

5. Motivation  Service Models and Mechanisms of IETF  Integrated Services/Resource Reservation Protocol (RSVP) Model  Differentiated Services (DS) model  Multi Protocol Label Switching (MPLS)  Traffic engineering  Constrained Based Protocol  How they differ from and relate to each other?  Which system they fit? 5

6. Outline 1. IntServ/RSVP 2. DiffServ 3. MPLS 4. Traffic Engineering/CBR 5. Comparison of ATM Networks 6. Conclusion 7. Q&A 6

7. IntServ/RSVP  Resource reservation  For real-time service, before data are transmitted,  Apps must first set up paths and reserve resources  Service classes  Guaranteed service: fixed delay bound  Controlled-load service: reliable and enhanced best-ef fort service 7

8. IntServ/RSVP 8

9.  Components  Signaling protocol  Admission control routine  Classifier  Packet scheduler 9

10. IntServ/RSVP  Limitations  Scalability State information is proportional to amount of flow  Overhead on routers Should have all components; RSVP, admission control, MF c lassification, packet scheduling  Ubiquitous deployment is required 10

11. DiffServ  Motivation  The difficulty in implementing and deploying IntServ and R SVP  DS is essentially a relative-priority scheme  Using DS fields of packet header to indicate service classes  Process complexity  core network → edge network  More scalable 11

12. DiffServ  Customer  Have a Service Level Agreement (SLA) with its ISP Which service class to provide?  Mark the DS field according to the service class  Router  Classification, policing, shaping, scheduling occur at only in gress routers  Domain  DF field is remarked by SLA between the domains 12

13. DiffServ  Possible services of DiffServ  Premium service For low-delay and low-jitter service  Assured service For better reliability than best-effort service  Olympic service gold, silver, and bronze, with decreasing quality  Differences between DiffServ and IntServ  State information is proportional to the number of classes  Process complexity is only at the edge router 13

14. DiffServ  Assured Service  Provide reliable services even in times of network con gestion  Be implemented as follows: Classification and policing are done at the ingress routers All packets, in and out, are put into an AQ The queue is managed by a RED or RIO 14

15. DiffServ  Premium Service  SLA specifies a peak bit-rate  Provide low delay and low jitter e.g.) Internet Telephony, Video Conferencing, VPN  More expensive, But more prior than Assured Service  Be implemented as follows: Use P-bit on DS field If P-bit is on, the packet goes to Premium Queue Uneven distribution of traffic may cause a problem 15

16. DiffServ  Service Allocation in Customer Domains  How to decide services; Assured or Premium Service Each host makes its own decision Bandwidth Broker(BB) makes decision  Resource Allocations in ISP Domains  How boundary routers handle incoming traffic Static SLAs, Manually configured. Dynamic SLA, RSVP 16

17. DiffServ 17

18. DiffServ 18

19. DiffServ  Requirements on Routers  Edge router: MF classifications, marking, and shaping  ISP ingress router: policing, re-marking  ISP egress routers: re-shaping  BA classification, Assured Queue, Premium Queue  For dynamic SLA, BB at the customer domain

20. MPLS  MPLS: Multi Protocol Label Switching  Incoming packets are assigned a “label” by edge router  Packets are routed according to the label  By a label switch router(LSR)  The path a packet traverses is called label switched path(LSP)  Network protocol independent

21. MPLS

22.  MPLS is strategically significant  Provides faster packet classification and forwarding  Provides an efficient tunneling mechanism without encryption  Moves processing to edge routers Core did forwarding only, Scalable

23. Traffic Engineering/CBR  Motivation  IntServ/RSVP and DiffServ has degradation of performance when traffic load is heavy.  Major goal  Provide efficient and reliable network operation  Optimize network resource utilization To efficiently manage bandwidth resources  Optimize traffic performance To enhance QoS of traffic stream

24. Traffic Engineering/CBR  The factor of Congestion  Lack of network resource Upgrade infrastructure  Uneven distribution Because of Shortest Path problem Constraint Based Routing Automatically Traffic Engineering

25. Traffic Engineering/CBR  Shortest Path Problem

26. Traffic Engineering/CBR  Constrained Based Routing  Compute QoS route May select longest lightly-loaded path rather than heavily shortest path  Improve network resource utilization

27. Traffic Engineering/CBR  Distribution of Link State Information  Needs of link available bandwidth, buffer space information  Link state advertisement of OSPF, IS-IS  Route Computation  Various algorithm for distinct metrics; cost, hop-count, bandwidth, reliability, delay, jitter Let d(i, j) be a metric for link (i, j). For any path P = (i, j, k, …,l, m), metric d is: Additive if d(P) = d(i, j) + d(j, k) + … + d(l, m) e.g.) delay, jitter, cost and hop-count Multiplicative if d(P) = d(i, j) * d(j, k) * … * d(l, m) e.g.) reliability (1-loss rate) Concave if d(P) = min{d(i, j), d(j, k), …, d(l, m)} e.g.) bandwidth

28. Traffic Engineering/CBR  Pros  Meeting the needs for QoS requirement of flows  Improved network utilization  Cons  Increased communication and computation overhead  Increased routing table size  Longer path may consume more resources  Potential routing instability

29. Traffic Engineering/CBR  The Position of CBR  DiffServ: Not for replacing, but helping DiffServ  RSVP: Independent with RSVP CBR just determines the path of RSVP messages  MPLS MPLS: forwarding scheme, CBR: routing scheme Work together for traffic engineering

30. Comparison of ATM Networks  ATM Network  Use Virtual Circuit Switching  Pros Fast Provide QoS  Cons ATM cell header overhead Switch can not work at the boundary of network  ATM Network with DiffServ or MPLS  Provide QoS on the router network  Reduce ATM cell header overhead

31. Conclusion  QoS is hotly debated issue  Fibers and WDM will make bandwidth so abundant and cheap QoS will be automatically delivered  However, New application will be invented to consume it Thus, Mechanism will be needed to provide QoS  Many mechanisms are provided but they not solve QoS problem  There is little hope for success  But, Way to go

32. Q&A