4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 1
Qos Management for VOIP Networks with Edge-to-Edge Admission Control
報告者 :
R93922106 莊萬慶 R93922073 王文廷 R93922127 張尚斌 R93525018 田謹維
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 2
Reference
• K. Mase, Y. Toyama, A.A. Bilhaj, Y. Suda, "QoS management for VoIP networks with edge-to-edge admission control", in:
Proceedings GLOBECOM 2001, vol. 4, 2001, pp. 2556 -2560.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 3Outline
• Motivation • Introduction• VoIP Network Management • Voice Quality Evaluation
• Edge-to-Edge Admission Control • Network Dimensioning
• Performance Evaluation • Conclusion
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 4Motivation
• VoIP 目前是在 Internet 上最吸引人的一項 Service 。• VoIP 在 Internet 上有許多的 Application 。 • 因為 VoIP 通訊便宜,所以現在許多人開
始使用 VoIP 的服務了,相對的 VoIP 就必
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 5
Introduction
• If a new call is accepted without a particular limit, QoS for calls in progress may be
degraded below an acceptable level,
because total bandwidth required for the calls exceeds the network capacity .
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 6
Introduction(Cont.)
• A mechanism called call admission control
is necessary to reject a new call when
enough network spare capacity is not available.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 7
Introduction(Cont.)
• Traditionally, the Internet has provided the
best effort services, and has not supported call admission control.
• However, admission control is necessary
for guaranteeing QoS for real-time
applications ( 如 : telephone service in the Internet).
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 8
Introduction(Cont.)
• Edge-to-edge measurement based
admission control (EMBAC), 它使用了
edge-to-edge probe flow and QoS
measurement to ensure spare capacity for the new flow. This method neither uses
hop-by-hop signaling, nor requires any additional functionality for routers in the backbone network.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 9
Introduction(Cont.)
• EMBAC 在 various network conditions 下,使 用 call admission control 來讓 both directional voice flows 的 packet loss rates 維持在 a pre-determined value 之內 .
• The results of voice quality evaluation is used to analyze possible problems, and if necessary to change parameters for admission control.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 10Outline
• Motivation • Introduction• VoIP Network Management
• Voice Quality Evaluation
• Edge-to-Edge Admission Control • Network Dimensioning
• Performance Evaluation • Conclusion
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 11
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 12
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 13
VOIP NETWORK MANAGEMENT(3)
• A VoIP network is designed to satisfy
requirements such as allowed budget and voice quality objectives. While VoIP
network is used, test calls are periodically generated between a set of PBX pairs, and voice quality and network-level QoS such as packet loss rates are measured for the test calls.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 14
VOIP NETWORK MANAGEMENT(4)
• As the results, problems in voice level QoS as well as network level QoS are identified. These problems are, then, analyzed and
fixed through admission control
optimization, network optimization, or fault and error recovery, depending on the
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 15
VOIP NETWORK MANAGEMENT(5)
• The typical admission control parameters include:
(1) average packet lost rate for VoIP flows. (2) admission threshold.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 16Outline
• Motivation • Introduction• VoIP Network Management
• Voice Quality Evaluation
• Edge-to-Edge Admission Control • Network Dimensioning
• Performance Evaluation • Conclusion
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 17
Voice Quality Evaluation(1/6)
• Measurement(1/2)
– Generate test call periodically
– Each test call undergoes the same admission control as ordinary calls do
– Once a test call is established, the artificial voice generation device attached to the call-originating PBX (Device A) sends artificial voice to the voice quality evaluation device attached to the
call-terminating PBX (Device B) through the forward VoIP path.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 18
Voice Quality Evaluation(2/6)
• Measurement(2/2)
– Device B calculates instantaneous MOS values as well as the average MOS by comparing the original artificial voice signal and the received voice signal.
– The VoIP gateway at the call-terminating PBX monitors and measures packet loss rate for the test call.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 19
Voice Quality Evaluation(3/6)
• Holding time for a test call is an important design parameter
• The shorter holding time is desirable to minimize increase in network traffic load, while it should be long enough to assure reliability in MOS evaluation
• G723.1 coding and enhanced PSQM algorithm are used
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 20
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 21
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 22
Voice Quality Evaluation(6/6)
• MOS 2 is a critical value for users to notice voice quality degradation.
• packet loss rate 2% is tolerable based on the measurement results of 20 or 60 sec
measurement time
• From these observations, 20 sec is a good candidate to obtain reliable voice quality evaluation.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 23Outline
• Motivation • Introduction• VoIP Network Management • Voice Quality Evaluation
• Edge-to-Edge Admission Control
• Network Dimensioning • Performance Evaluation • Conclusion
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 24
Edge-to-Edge Admission
Control(1/6)
• End node A (source) to end node B (destination) through a selected path
• Node B is in charge of the admission test and judges whether to be able to accept the flow from node A to node B or not.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 25Edge-to-Edge Admission
Control(2/6)
Endpoint(Node O) Endpoint(Node T) Probe Request Probe Connect Voice Exchange Release Complete4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 26
Edge-to-Edge Admission
Control(3/6)
• The probe request is sent from the originating node (node O) to the call-terminating node (node T)
• Node O and node T may become a source or destination of the probe packet flows, as
mentioned before
• The probe request activates node T to initiate the probing and measurement operation.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 27
Edge-to-Edge Admission
Control(4/6)
• Following the probe request transmission and reception, probe packet flows are
carried in both direction and packet loss rate measurements are conducted at the both end nodes.
• Node O measures the packet loss rate for the probe flow from node T to node O, and conducts admission test.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 28
Edge-to-Edge Admission
Control(5/6)
• If the result of the admission test is success, node O transmit setup signal to node T.
• If it is failure, node O terminates the call setup.
• In parallel, node T measures the packet loss rate for the probe flow from node O to node T, and conducts admission test.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 29
Edge-to-Edge Admission
Control(6/6)
• If the result of the admission test is success, node T proceeds to transmit connect signal to node O, responding the setup signal sent from node O.
• If it is failure, node T will reject setup request from node O.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 30Outline
• Motivation • Introduction• VoIP Network Management • Voice Quality Evaluation
• Edge-to-Edge Admission Control
• Network Dimensioning
• Performance Evaluation • Conclusion
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 31
V.Network Dimensioning
• For dimensioning the network we need assume some parameters
• Between source and destination ,There have “d“ links
d links
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 32
Traffic Matrix
• Location-to-Location VoIP traffic demands are represented by traffic matrix a[i,j]
• “i” represents the source
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 33
Parameter Assuming
Parameter Explain
B the edge-to-edge blocking probability is no more than this pre-determined value “B”
F Edge-to-edge peak pocket loss rate is no more than a pre-determined value “F”
P Represents the percentage of time during which speech is present
Freezeout
fraction (average number of frozen out calls)/(average number of active calls)
k The number of the VoIP calls
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 34
The peak values
• Blocking probability no more than B/d • Freezeout fraction no more than F/d
• The Freezeout fraction is the “upperbound” for the packet loss rate
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 35Outline
• Motivation • Introduction• VoIP Network Management • Voice Quality Evaluation
• Edge-to-Edge Admission Control • Network Dimensioning
• Performance Evaluation
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 36
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 37
Simulation Model and
Assumptions(1)
• For simplicity ,we ignore the amount of signaling flows ,because it is not significant compared with that of voice flows.
• A PBX is modeled as a switch accommodating – infinite number of subscriber lines
– infinite number of outgoing and incoming trunks
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 38
Simulation Model and
Assumptions(2)
• Calls originate
– according to Poisson distribution between a pair of call-originating and terminating
locations
• Call holding time
– base on exponential distribution with the average three minutes.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 39
Simulation Model and
Assumptions(3)
• We assume that
– blocking probability target B is 3% – freezeout target F is 1.5%.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 40
Scenario (1)
• A bottleneck may occur in the network due to traffic forecast error.
• We select a link in the middle of the
network as the bottleneck link, and decrease the capacity from the initial size.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 41
Scenario (2)
• We use blocking probability and the peak packet loss rate for two seconds interval as the performance parameter.
• We consider three cases
– No Admission Control – Admission threshold 2% – admission thresholds 10%
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 42
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 44
with admission control
• the peak packet loss rate is remarkably
improved at the cost of acceptable increase in blocking probability, depending on the given admission thresholds.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 45
without admission control
• blocking probability is always 0, and peak packet loss rate is beyond acceptable level even without capacity reduction, and
increases as the capacity reduction increases.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011 2005/6/14 46Outline
• Motivation • Introduction• VoIP Network Management • Voice Quality Evaluation
• Edge-to-Edge Admission Control • Network Dimensioning
• Performance Evaluation
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 47
CONCLUSION(1)
• Admission control works well to control packet loss rate under given network
conditions.
• We need to properly set the admission thresholds for each end node pair
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 48
CONCLUSION(2)
• The relation of packet loss rate and the
admission threshold is not obvious and it is not easy to analytically find the optimal
admission threshold.
• Feedback control based on voice quality
and packet loss measurements may be used to dynamically adjust the admission
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 49
REFERENCES
• [1] B. Li, M. Hamdi, D. Jiang. Y. T. Hou, and X. Cao, “QoS-enabled voice
support in the next-generation Internet: Issues,, existing approaches and challenges,” IEEE Communications Magazine, Vol.38, No.4, April, 2000.
• [2] L. Breslau, E. Q. Knightly, S. Shenker, I. Stoica, and H. Zhang, “Endpoint
admission control: architectural issues and performance,” pp. 57-69, SIGCOMM’00, 2000.
• [3] F. Borgonovo, A. Capone, L. Fratta, M. Marchese, and C. Petrioli, “PCP: A
bandwidth guaranteed transport services for IP networks,” ICC’ 99, pp. 1999.
• [4] G. Bianchi, A. Capone, C. Petrioli, “Throughput analysis of end-to-end
measurement-based admission control in IP,” INFOCOM 2000, 2000.
• [5] V. E.lek, G. Karlsson, and R. Ronngren, “Admission control based on
end-to-end measurement,” INFOCOM 2000, 2000.
• [6] M. Schwartz, K. Mase, and D. R. Smith, “Priority channel assignment in
tandem DSI,” IEEE Trans. on Communications, Vol.Com-28. No.10, 1980.
• [7] http://www.radcom-inc.com/products/internetsim.htm.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 50
• Each packet has 40 bytes overhead
– 20 bytes IP packet header – 8 bytes UDP header,
– and 12bytes RTP header
• The maximum length of a packet is 60 bytes.
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 51
• A VoIP gateway and a router are modeled as a queuing system.
• Voice flows and probe flows are given individual classes and their own queues.
• As mentioned in ,voice flow is given high priority Ⅳ in packet scheduling than probe flows.
• Specifically, non-preemptive priority scheduling is used.
• Buffer size is 40 packets for voice flows and 20 packets for probe flows for each output link .
4
1
5
2
0011 0010 1010 1101 0001 0100 1011
2005/6/14 52
• One packet is generated every 20 ms during active periods for each call.
• Thus, the maximum rate for a VoIP call, w, mentioned in Sec. , is 24 kbps.Ⅴ
• We assume probe calls have one second duration.
• A size of a probe packet is always 60 bytes. • The admission threshold is set to 10 %.