1. 1 Call Admission Control Carey Williamson Department of Computer Science University of Calgary

2. 2 The CAC Function: Everything you ever wanted to know but were afraid to ask Carey Williamson Department of Computer Science University of Saskatchewan

3. 3 Introduction u The purpose of an admission control algorithm is to decide, at the time of call arrival, whether or not a new call should be admitted into the network u A new call is admitted if and only if its Quality of Service (QOS) constraints can be satisfied without jeapordizing the QOS constraints of existing calls in the network

4. 4 Call Admission Control u Admission control decision is made using a traffic descriptor that specifies traffic characteristics and QOS requirements u Traffic characteristics: –peak cell rate (PCR), sustained cell rate (SCR), maximum burst size (MBS),... u QOS requirements: –tolerable cell loss, cell delay, delay variation

5. 5 Issues u Want to make efficient use of the network (i.e., accommodate as many calls as possible, and maintain a reasonably high level of network utilization) u Want to guarantee quality of service for all calls that get into the network u Tradeoff: can’t always have both!

6. 6 Why is it Difficult? u ATM is completely based on the idea of statistical multiplexing of VBR sources u No assignment of specific slots to users, but statistical assignment of capacity based on expected traffic characteristics u Providing guarantees requires conservatism u High utilization requires aggressiveness

7. 7 Why is it Difficult? (Cont’d) u Typical traffic sources are bursty u Some traffic sources are VERY bursty u Traffic can be highly unpredictable u Accurate traffic descriptors may not be known in advance u Traffic may not conform to its descriptor

8. Least Understood Most Understood Traffic Characterization

9. Least Understood Most Understood Voice Traffic Characterization

10. Least Understood Most Understood Voice CBR video Traffic Characterization

11. Least Understood Most Understood Voice CBR video Packet data Traffic Characterization

12. Least Understood Most Understood Voice CBR video Packet data Traffic Characterization Image

13. Least Understood Most Understood Voice CBR video Packet data Image VBR video Traffic Characterization

14. 14 Multiplexing u Two basic approaches u Deterministic multiplexing u Statistical multiplexing

15. 15 Deterministic Multiplexing u The traditional means of bandwidth allocation in telecommunications networks u Each traffic type has an inherent bit rate (e.g., voice traffic = 64 kilobits per second) u Allocate precisely that bandwidth for each call, for the duration of the call

16. 16 Deterministic Multiplexing (Cont’d) u Advantages: –Simple –Works great for CBR traffic (PCR = SCR) u Disadvantages: –Inefficient for VBR traffic (PCR !=SCR) u Allocating PCR can waste lots of capacity

17. Deterministic versus Statistical Multiplexing Bit rate Source 1: peak 12 Mbps, mean 8 Mbps

18. Deterministic versus Statistical Multiplexing Bit rate 12 Mbps

19. Deterministic versus Statistical Multiplexing Bit rate Source 2: peak 10 Mbps, mean 6 Mbps

20. Deterministic versus Statistical Multiplexing Bit rate 22 Mbps (12 + 10)

21. Deterministic versus Statistical Multiplexing Bit rate 22 Mbps (12 + 10) Average utilization will be 14/22 = 64%

22. Deterministic versus Statistical Multiplexing Bit rate

23. Deterministic versus Statistical Multiplexing Bit rate

24. Deterministic versus Statistical Multiplexing Bit rate

25. Deterministic versus Statistical Multiplexing Bit rate

26. Deterministic versus Statistical Multiplexing Bit rate

27. Deterministic versus Statistical Multiplexing Bit rate

28. Deterministic versus Statistical Multiplexing Bit rate

29. Deterministic versus Statistical Multiplexing Bit rate

30. Deterministic versus Statistical Multiplexing Bit rate

31. Deterministic versus Statistical Multiplexing Bit rate Bandwidth saving with Statistical Multiplexing

32. 32 Statistical Multiplexing u Basic idea: ‘‘pack in’’ more than would be able to fit with deterministic multiplexing u Takes advantage of the variable bit rate bursty nature of traffic u Not all traffic sources will need their peak rate at the same time (hopefully) u Peaks and valleys should balance out

33. 33 Statistical Multiplexing (Cont’d) u Advantages: –More calls can fit in the network –Increases utilization, efficiency of network –Statistical gain can be significant u Disadvantages: –QOS is hard to guarantee (100% guarantee) u Always an element of risk, however slight

34. 34 Simple CAC Schemes int CAC_Function(TrafficDescriptor *TD) { return( YES ); }

35. 35 Simple CAC Schemes int CAC_Function(TrafficDescriptor *TD) { return( YES ); }

36. 36 Simple CAC Schemes (Cont’d) int CAC_Function(TrafficDescriptor *TD) { return( NO ); }

37. 37 Simple CAC Schemes (Cont’d) int CAC_Function(TrafficDescriptor *TD) { return( NO ); }

38. 38 Possible CAC Schemes u Peak rate allocation u Mean rate allocation u (Peak + Mean) / 2 u Virtual Bandwidth [Murase 90] u Schedulable Region [Lazar 91] u Effective Bandwidth [Elwalid 93]

39. 39 Peak Rate Allocation u Allocate the peak cell rate for the source u Same as Deterministic Multiplexing u Guarantees that no cell loss occurs u Guarantees that bandwidth is wasted if source is at all bursty (peak > mean) u The amount of wasted bandwidth depends on the peak-to-mean ratio

40. 40 Mean Rate Allocation u Allocate bandwidth based on the mean rate (SCR) u By definition, this is adequate over a long enough time duration u Drawback is the delay for traffic bursts u May not be enough capacity to handle bursts within a tolerable delay

41. 41 (Peak + Mean) / 2 u Peak rate is the most that is needed u Mean rate is the least that is needed u ‘‘Correct’’ allocation must be in between u But where is the real question! u (Peak + Mean) / 2 is one guess u Suitability depends on characteristics of source (e.g., time spent at or near each)

42. 42 Can you do better? u Of course! u Effective Bandwidth: [Elwalid 93] u Virtual Bandwidth: [Murase 90] u Schedulable Region: [Lazar 91] u We’ll look at some of these in more detail in just a moment...

43. 43 Summary u Call Admission Control is one of the most difficult problems to deal with in ATM networks u Difficult problem, no standard solution u Lots of research activity u Impossible to find a single ‘‘best’’ answer