2001/09/26 Chin-Kai Wu, CS, NTHU 1
Resource Allocation and Resource Allocation and Management in DiffServ Management in DiffServ Networks for IP TelephonyNetworks for IP Telephony
Maarten Büchli, Danny De Vleeschauwer, Jan Janssen,
Annelies Van Moffaert, Guido H. Petit
11th International workshop on Network and Operating Systems support for digital audio and video, June 25 - 26, 2001, Port Jefferson, NY USA
2001/09/26 Chin-Kai Wu, CS, NTHU 2
OutlineOutline Introduction
Voice Traffic Model Quality of Service
End-To-End Delay Analysis Mouth-to-ear delay Queuing delay
Simulation
Results
2001/09/26 Chin-Kai Wu, CS, NTHU 3
IntroductionIntroduction In DiffServ, a scheduler is needed in order to
guarantee throughput and delay to the different aggregate flows in the system.
The class-based Weighted Fail Queuing (WFQ) is considered. Hence, dimensioning the bandwidth for the voice traffic is therefore of great concern.
The paper shows how to set the weights of a WFQ scheduler in a DiffServ router such that a certain delay bound is met.
2001/09/26 Chin-Kai Wu, CS, NTHU 4
Voice Traffic ModelVoice Traffic Model
Steps
Packet Size
Analog Voice
Sample Quantize Encode Packitize
M: voice packet sizeRcod: codec bit rateTpack: packetization delaySOH: header overhead
2001/09/26 Chin-Kai Wu, CS, NTHU 5
CBR voice flows is well modeled by a Poisson process, and cannot become more bursty than it.
Voice sources that use VAD (Voice Activity Detection) are much harder to characterize.
2001/09/26 Chin-Kai Wu, CS, NTHU 6
Quality of ServiceQuality of Service
2001/09/26 Chin-Kai Wu, CS, NTHU 7
Capacity assigned to voice traffic
Queue load
Cvoice: capacity of voice trafficφvoice: weight of voice traffici: number of queues
ρ: queue loadBvoice: average bit rate of voice flows
2001/09/26 Chin-Kai Wu, CS, NTHU 8
Mouth-to-ear delayMouth-to-ear delay
Deterministic part Tpack, packetization Tser, serialization Tprop, propagation Tdejitter, dejittering Toth, other (encoding, decoding etc.)
Stochastic part Tqueue, queuing delay
2001/09/26 Chin-Kai Wu, CS, NTHU 9
Only interested in the maximum queuing delay. For this delay, a reasonable quantile can be used since this is the fraction of packets that arrive in time.
The queuing delay has to be compensated for in the dejittering buffer of the receiver. Hence, the dejittering delay should be chosen equal to a quantile of the queuing delay.
When perfect dejittering is used, the queuing and dejittering delay of each packet is equal to the maximum queuing delay.
2001/09/26 Chin-Kai Wu, CS, NTHU 10
Delay distribution
2001/09/26 Chin-Kai Wu, CS, NTHU 11
Queuing delayQueuing delay
R: reserved rateb: max burst sizeM: max voice packet sizeN: number of hopsMTUi: MTU at node iCi
link: link capacity at node i
Ri: reserved rate at node ibi: max burst size at node i
2001/09/26 Chin-Kai Wu, CS, NTHU 12
Simulation ScenarioSimulation Scenario
2001/09/26 Chin-Kai Wu, CS, NTHU 13
Simulation ResultsSimulation Results
2001/09/26 Chin-Kai Wu, CS, NTHU 14
2001/09/26 Chin-Kai Wu, CS, NTHU 15
ResultsResults
2001/09/26 Chin-Kai Wu, CS, NTHU 16