Transcript D-4 - TMCnet

QoS for VoIP:
Measuring its Effectiveness
Bob Massad
Telchemy®, Inc
www.telchemy.com
[email protected]
2/07/03
Telchemy 2003
1
Agenda
• QoS Definitions
• The need for QoS
• Defining QoS Effectiveness as QoE
• Measuring QoS Effectiveness
• Concluding Remarks
2/07/03
Telchemy 2003
2
What is QoS?
• “… providing consistent, predictable
data delivery service.” (QoS Forum)
• “…the … assurance that requirements
can be satisfied.” (QoS Forum)
• “…service requirements to be met …
while transporting a flow.” (RFC 2386)
• In short: managing bandwidth, priority
and queues – not QUALITY
2/07/03
Telchemy 2003
3
QoS Mechanisms
•
•
•
•
•
•
•
•
ToS – Type of Service
802.1p – Layer 2
IntServ – Integrated Services
DiffServ – Differentiated Services
RSVP – Resource Reservation
CoS – Class of Service
CAR – Committed Access Rate
CIR – Committed Information Rate
2/07/03
Telchemy 2003
4
2 Interesting QoS Services
• Integrated Services (RFC 1633 , RSVP 2205)
– Dynamic per hop RSVP request for a “quantity”
– Per node parameters signaled per flow
– Two currently defined services:
• Guaranteed service
• Controlled-load service
• Differentiated Services (RFCs 2475 & 3086)
– Uses TOS byte (DSCP) to classify packets
– DSCP used to “treat” (PHB) packets per hop e2e
– 3 PHBs:
• EF - low latency, jitter, loss - Virtual Leased Line
• AF - classifies precedence levels: G, B, S (Olympic)
• DE – Default/Best Effort
2/07/03
Telchemy 2003
5
Why Internets need QoS?
• IP provides few packet services
– Routing/Addressing
– Fragmentation and Reassembly
• TCP provides reliability but…
– No throughput management
– No latency management
• UDP only provides port multiplexing
• Internets are shared media
• New Apps on the Horizon – VoIP…
2/07/03
Telchemy 2003
6
From: Ryan Koga, [email protected] and
Sean McCreary, [email protected]
2/07/03
Telchemy 2003
7
Heterogeneous Apps
2/07/03
Telchemy 2003
8
QoS Effectiveness for VoIP
• Different than typical Internet apps
– Typical Apps
• Email
• File transfer
• HTTP/Web
– Measured via Thru-put, Response Time, Retries…
•
•
•
•
VoIP Uses UDP
VoIP Real-time
VoIP Sense/Experience oriented
Measured via Quality of Experience (QoE)
2/07/03
Telchemy 2003
9
What is QoE for VoIP?
Listening Quality (LQ) - call clarity
and presence/absence of disturbing
effects
Conversational Quality (CQ) –
degree of support or interference with
dialog, i.e. conversation.
2/07/03
Telchemy 2003
10
Factors in VoIP Call Quality
IP
Network
CODEC
type
2/07/03
Local
Loop
Packet Loss, Delay, Jitter
Telchemy 2003
11
The short stories
• Jitter
– Jitter Buffer moderates effects
• Turns Jitter into small Delay
• Turns problematic delay into packet loss
• Delay
– Not indicative of LQ unless excessive
– More indicative of CQ
– Excessive delay becomes packet loss
• Leaving packet loss…
2/07/03
Telchemy 2003
12
The Key VoIP Problem:
Packet Loss/Discards
Gateway
Jitter Buffer
Gateway
CODEC
Lost Packets
2/07/03
Telchemy 2003
Jitter Buffer
Discards
13
Packet Loss is Bursty
2/07/03
Telchemy 2003
14
Packet Loss Concealment
For consecutive packets lost, same packet repeated
multiple times !
Worse if long or multiple voice frames per packet!
2 frames lost
Same frame
repeated 3
times
2/07/03
Telchemy 2003
15
Expect more
packet loss in UDP
apps!
One study showed
mean packets lost
in burst = 6.9 !
…Another shows
when 1 packet is
lost the probability
increases that
next packet will be
lost too!
1% of loss bursts
can account for
50% of all losses!
2/07/03
Telchemy 2003
16
End-user Perception
• Transitional Quality Recognition
– From bad to good
– From good to bad
• Recency
– Short term auditory memory
– Length of time since last impairment
2/07/03
Telchemy 2003
17
“Recency” effect
Start
Of call
60 seconds
End
Of call
MOS 3.8
MOS 3.3
MOS 3.2
“Good” 4.3MOS
2/07/03
“Bad” 1.8 MOS
(3dB SNR)
Telchemy 2003
Source AT&T
T1A1.7/98-031
18
Enhanced E-Model
Packet
Loss
Jitter
Codec
type
Delay
Loss
Model
Jitter
Model
Ie
Network
R Factor
Burst
model
User
Recency R Factor
Codec
Model
model
Delay
model
Extensions incorporated
in ETSI Standard TS
101329-5 Annex E
2/07/03
Telchemy 2003
19
Comparison with E Model
Comparison of VQmon and E-Model
High Bursts
4.5
MOS
4
Vqmon
3.5
Subjective
3
E-Model
2.5
2
1
4
7
1,4
1,7
4,7
1,4,7
Impairment Locations
2/07/03
Telchemy 2003
20
Call Quality Monitoring
QoS
Domain
SLA
monitor
IP Network
QoS
Domain
IP Network
Jitter buffer
simulator
VoIP
End-Point
CODEC
VQmon
VQmon
2/07/03
Telchemy 2003
21
Conclusions
• Control of QoS for VoIP involves control of
key parameters within IP Net.
• Real-time monitoring of Call Quality/QoE
provides a comprehensive means of
measuring the effectiveness of network QoS
parameters and points to remedial actions.
• Non-intrusive objective Measurement based
on Enhanced E-Model provides an accurate
metric of instantaneous Call Quality/QoE.
• It’s a similar case for Video!
2/07/03
Telchemy 2003
22
About Telchemy
•
•
•
•
•
Founded - Late ‘99
Location - Atlanta, GA
Focus - embedded tools to optimize V²oIP
Approach - open software architecture
First products - VQmon®/EP and /SA lightweight embedded call quality
monitoring agents for integration in VoIP
endpoints and management devices.
2/07/03
Telchemy 2003
23