Transcript Performance Analysis for VoIP System

Performance Analysis for
VoIP System
Members
R94922009 周宜穎
R94922020 吳鴻鑫
R94922064 張嘉輔
Outline
What is Performance
Performance Bound
How to analyze Performance
Some Performance Analysis Exmaple
What is Performance ?
[10]
There are numerous factors that affect the
performance assessments.
– Human factors
– Device factors
– Network factors
Human factors -Audiovisual Quality
Assessment Metrics
Subjective quality assessment - MOS
Objective quality assessment
–
–
–
–
–
Signal-to-Noise Ratio (SNR)
Mean Square Error (MSE)
Perceptual Analysis Measurement System (PAMS)
Perceptual Evaluation of Speech Quality (PESQ)
E -model
E-model
R-scale ( 0 to 100 ) <=> MOS rankings and User
Satisfaction
Human factors
Voice Quality Classes
Device factors
Essential devices such as
–
–
–
–
–
–
–
–
–
–
VoIP endpoints
Gateways
MCUs (Multipoint Control Units )
Routers
Firewalls
NATs (Network Address Translators )
Modems
Operating System
Processor
memory
Network factors
Network congestion
Link failures
Routing instabilities
Competing traffic
General Measruement :
– Delay
– Jitter
– Packet loss
The Performance Standard
Delay
– Good (0ms-150ms)
– Acceptable (150ms-300ms)
– Poor (> 300ms)
Jitter
– Good (0ms-20ms)
– Acceptable (20ms-50ms)
– Poor (> 50ms).
Loss
– Good (0%-0.5%)
– Acceptable (0.5%-1.5%)
– Poor (> 1:5%)
E-model
ITU-T recommendation
Well established computational model
Using Transmission parameters to predict
the quality
We can get the basic Performance
Standard by through the model
Basic formula for the E-model
R-value = Ro - Is - Id - Ie + A
– Ro
the basic signal-to-noise ratio based on sender and receiver
loudness ratings and the circuit and room noise
– Is
the sum of real-time or simultaneous speech transmission impairments,e.g.
loudness levels, sidetone and PCM quantizing distortion
– Id
the sum of delay impairments relative to the speech signal, e.g., talker echo,
listener echo and absolute delay
– Ie
the equipment impairment factor for special equipment, e.g., low bit-rate
coding (determined subjectively for each codec and for each % packet loss
and documented in ITU-T Recommendation G.113)
– A
the advantage factor adds to the total and improves the R-value for new
services.
Estimating the R
R = (Ro − Is) − Id − Ie + A
Ro , Is
– do not depend on network environment
Id
– This the Argument of Delay
Ie
– It mostly affect by codec and packet loss
A
– Additional adjust argument ,not considered in general
Estimating Id and Ie
Id = Idte + Idle + Idd
– Idte -Talker echo delay
– Idle - Listener echo delay
– Idd - Long delay
Ie
– It base on codec, but packet loss affect can be
emulated as a function
Estimating Id and Ie
The distortion as a function of packet loss
also depends on whether or not PLC
(Packet Loss Concealment)
increases 4 units for codecs with PLC (in
the R scale per 1% packet loss)
10 units for codecs without PLC
Curve Diagram
Test Setup
Using 9 scenarios to
test 27 possibilities
Using NISTnet
network emulator
(http://snad.ncsl.nist.g
ov/itg/nistnet/)
create the various
network health
scenarios
MOS Vs Delay
MOS Vs Jitter
MOS Vs Loss
Normalized
Each unit in the normalized scale corresponds to
delay : 150ms
jitter : 20ms
loss : 0.5%.
The Conclusion about Performance
bounds
We show that end-user perception of
audiovisual quality is more sensitive to the
variations in end-to-end jitter than to
variations in delay or loss
We get a simple standard about the
Performance to estimate Performance
How to Analyze Performance
Thinking about two topic
– Measurement
– Network Condition
Measurement mean the analysis model
that estimate key parameters
Of course, it is the way to compute delay,
jitter ,packet loss
Two Measurement
[7],[8],[9]
There are two methods in performance
measurement
passive measurement
– records and analyzes existing traffic.
active measurement
– Inject sample packets into the network.
Introduce a simple Measure
Measurement Method in LAN
sends sequences of UDP packets to
unlikely values of destination port numbers
(larger than 30,000)
This causes the destination host’s UDP
module to generate an ICMP port
unreachable error when the datagram
arrives
ICMP
TCP/UDP/IP 協定若有錯誤情形發生時，會
利用 Internet Control Message Protocol
（ICMP）協定來送錯誤訊息 。
在 ICMP 的 type 中，目前約有 15 種
The ICMP echo mechanism should be
installed in host in the measurement
RTT of one sent packet
How to Compute?
Ti = Bi / v
+ Di /v
+ CL
+C
Ti − CL
=(Bi + Di) / v
+ C.
Keep estimating
one-way delay (T i )
T i = (Ri − Si) −Di / v −CL/2 − C/2
This calculation assumes that all delay
happens on the sending path.
J i,i+1 = (T i+1 − T i )
Packet loss = packetslost / packetssent
How about more complicated?
Precision timestamping
Queuing Model
Special Model for Protocol or device
Seem to Traffic Analysis!?
Ex: SIP Traffic Model
[11]
A model for SIP
Traffic
Two Sub Model
– IP Path Model
– SIP Finite State
Machine
FSH Notation
Q = State set
M = fixed number of sessions
C = the bottleneck transmission rate( bit/s)
R = total capacity of IP Path measure in packets
of D bits
rtt = round trip time measured in seconds
p = probability of 3xx Response
ps = successful probability of packet
transmission
Sample Computation
Call Dropping rate pcd
Enviroment condition for VoIP
performance
[4] , [5]
The aspects about VoIP Performance
Analysis
Protocols
– H.323 v.s. SIP
Network
– Ethernet network v.s. wireless LAN (WLAN)
network
Security for VoIP Communication
– VPN protocols : PPTP v.s. IPSec
Delay in Ethernet Network
Both SIP and H.323 incurred higher delays
in secure network-to-network environment.
SIP
H.323
Jitter in Ethernet Network
IPSec produced the highest jitter values
for both H.323 and SIP communications.
Jitter in Wireless-LAN
IPSec-based VoIP communications
generally incurred the highest jitter values.
Packet Loss Rates
IPSec and PPTP increased the packet
loss rate in both Ethernet and WLAN.
SIP
H.323
Performance in Satellite Network
Also provides IP-base data services
For remote region
As backup links
[1]
The purpose
The performance under
– Delay
– Random errors , burst errors
– Link loading
Two codecs
– 8 kb/s G.729
– 6.3/5.3 kb/s G.723.1
Test bed configuration
Baseline Tests
Bandwidth and bandwidth efficiency
Environment
– No background traffic
– No error
– Link delay set 270ms
– Run 15min with all 24 channel
Bandwidth Efficiency
5
5
Bandwidth
A single channel
Link Errors Tests
Random Error Tests and burst Error Tests
BERs (bit error rates) = BD/(B+GC)
– Burst length (B)
– Burst density (D)
– Gap length (G)
– Link capacity kb/s (C)
Random Error Tests
Burst Error Tests
Link Loading Tests
Environment
– With different link loading levels
– Link errors or not
– Packet loss
– Packet delay
Tests with an Error-Free Link
Tests with an Error-Free Link
Tests with Errors
Combine effect of both link loading and
link errors.
Error ↑,background traffic↓
link loading level↓
 link loading level can’t be predetermined
Impact of link failures on VoIP
performance
[3]
Three major causes of performance degradation
– network congestion
– link failures
– routing instabilities
Congestion is always negligible.
Link failures may be followed by long periods of
routing instability.
The goal is to study the impact of link failures on
VoIP performance.
Portion of the network topology
Solid arrows: primary path
Dashed arrows: alternative path used after
the failure
Impact of failures on data traffic
06:34 R1, R2, R5: link to R4 is down
06:35 R1, R2, R5: adjacency with R4 recovered
06:36~06:47 R4 is instable
Impact of failures on data traffic
06:48
06:59
07:17
07:36
R4 finally reboots
R4 builds its first routing table
R1, R2, R5: link to R4 is definitely up
an alternative path is chosen
Impact of failures on data traffic
the failure we observed in four phases
– 06:34 link is down, delay↑, few packet loss
– 06:36~06:47 router is instable, same delay,
packet loss↑
– 06:48~07:04 router reboots, no delay, packet
loss↑
– 07:05~07:17 router builds routing table,
delay↑, packet loss↑
Referencs
[1]Voice over IP Service and Performance in Satellite Network, IEEE
Communications Magazine
[2]Technique for Performance Improvement of VoIP Applications, IEEE MELECON
2002
[3]Impact of link failures on VoIP performance, ACM 1-58113-512-2/02/0005
[4]VoIP Performance Measure Using Qos Parameters , The Second International
Conference
[5]VoIP Performance Management, Internet Telephony Fall 2005
[6]Comparative Analysis of Traditional Telephone and VoIP System
[7]VoIP Performance on differenriate service
[8]Measuring Voice Readiness of Local Area Networks
[9]Experimental Investigation of the Relationship between IP Network Performances
and Speech Quality of VoIP
[10]Performance Measurement and Analysis of H.323 Traffic
[11] A Technique to Analyse Session Initiation Protocol Traffic
Appendix 1
Delay within the E-model
Id = Idte + Idle + Idd
– Idte -Talker echo delay
– Idle - Listener echo delay
– Idd - Long delay
The E-model delay measures
T – mean one-way delay
Ta – absolute delay
Tr – round-trip delay
Id can be computed by three argument
VoIP delay estimate
Drtcp – delay estimate from RTCP packets.
De – coding and packetization delay (at least as large
as packet size)
Dj – delay introduced by jitter buffer and decoder
Ds – send side’s access delay
Dr – receive side’s access delay
Delay measures transform
T = Drtcp + Dj + De + Dr
Tr = 2 * Drtcp + Dj + De
Ta = Drtcp + Dj + De + Dr + Ds
So the following importance is ….
How to estimate the Delay?
Estimation of Delays
Estimation of Ds and Dr defaulted to zero
Estimation of De
–
–
–
–
–
the length of a coded fram
the codec lookahead
the number of frames in the packet
the efficiency of the coder.
choosing best-case + 20% of the frame size would
be a reasonable estimate of encoding delay
Estimation of Delays
Estimation of Drtcp
Drtcp is the round-trip delay estimate
divided by 2.
Estimation of Dj
This is dependent on the VoIP gateways
jitter buffer and decoder.
A possible equation for Dj is:
Dj = min ( codec_frame_size + 0.9 *
RTP_jitter , 300 );
Appendix 2
Qos
Parameters of VoIP performance
and improvement techniques
[2] , [6] , [7]
End-to-End Delay
Jitter
Frame erasure
Out-of-order packet delay
End-to-End Delay
The delay from the mouth of speaker to
the ear of listener
Network delay
packet processing in both end system
packet processing in network device
propagation delay
Others (but leave out here)
speech processing
speech compression
speech packetization
Network delay
fixed part
In every network note (router) IP packets are
delayed
propagation delay
transmission delay
variable part
the time spent in queues of the network nodes on
the transmission path
Reduce network delay
fixed part
If the network and the transmission path are fixed
shorter IP packets
variable part
Some advanced queue-scheduling mechanisms
e.g. the IETF document RFC 2598
Using fragments time of long packets to send
Reduce jitter
employ a playout buffer
packet loss
Trade-off
playout time =1
additional delay
playout time =2
jitter absorption
Three main technique
fixed playout time static playout time
Adaptive adjusting of the playout time
during silence periods
Constantly adapting the playoit time for
each individual packet
Frame erasure
the packet does not arrive in time
is corrupted during the transmission through the
network
is dropped because of the network congestion
is lost because of a network malfunction
just arrives too late
Reduce frame erasure
FEC (Forward Error Correction)
need additional bandwidth and increases delays
because additional processing.
Loss concealment
be used independently or in the combination with
FEC
is effective only at low loss rate of a single frame
Out-of-order packet delay
occurs in the network with a complex
topology
Done in the jitter buffer
reordering (using RTP header)
elimination of jitter