Transcript 14Wilkins

Linking User Acceptance and
Network Performance
Miles Wilkins (BT)
P807 (JUPITER2)
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
IP-based Services
• And many more …
• The world has gone IP mad (not ATM)
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
But….
•
•
•
•
•
•
•
ATM can guarantee bandwidth / delay
IP services only “Best Effort”
QoS support being added by IETF and others
Real-time requirements (multimedia)
Not enough (reliable) bandwidth/delay on the Internet
Early use in corporate intranets
How do you support these applications?
– and protect other data flows?
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
What is ‘QoS’?
• Quality of Service
– variously defined by ITU-T (E.800) and others
• Objective
– Network measurements
• Subjective
– User’s perception and expectations
• “Constantly meeting customers expectations
in a service”
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
What is the relationship between user perceived
QoS and the actual network QoS?
I can’t hear
him very well.
Video
IP Network
Server
How was it
for you?
Multimedia
Collaboration
Introduce packet
loss & delay
What is a realistic range of loss / delay values?
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Determining ‘typical’ parameters
Intranet
• Measure traffic characteristics of applications
– NetMeeting, NetShow, Cisco IP/TV
• Generate similar test traffic
– with sequence numbers and timestamps
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Laboratory Subjective Tests
User
Video Server
or
Second User
Applications
- NetMeeting
- NetShow
- IP/TV
Tasks
- video clips
- editing/
discussion
Network Impairment
- Packet Loss
- Packet Burst Loss
- Packet Delay
- Packet Jitter
(perceived as loss/delay by user)
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Data Collection
- Questionnaire
- Interview
- video and screen
capture analysis
Main Interests & Results
•
•
•
•
Audio quality
Video quality
Overall quality
Acceptability
– (would you use this system again with this quality?)
• Quality and acceptability judgements were
affected by the amount of loss exhibited by a
network and by packet burst size
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Video Streaming Results
MO S
5
4
Burst
Siz e
3
1-2
6-7
9-10
2
1
0.5
1
4
7
% Loss
• Effect of packet loss & burst size on IP/TV
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Video Streaming Results
MOS
5
4
Quality
Overall
3
Video
Audio
2
1
Loss
Burst
0
0
1/2%
1/2%
1-2
6-7
Network Condition
1%
1-2
• Effect of packet loss & burst size on NetShow
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Conferencing Results
Video Ratings
Audio Ratings
• Audio quality most important and most disturbing
– (for some tasks)
• Rating drops between 50mS-120mS jitter
– caused by receiver buffer overflow (loss)?
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
The Next Experiments
• Field Trials
– video streaming and conferencing applications
– network impairment on source
• Video Streaming
– validate laboratory tests
– live video source (BBC News 24)
– NetShow & IP/TV
• Conferencing
– packet loss burst effect?
– NetMeeting
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Performability
• Taking results of user subjective test determine how
to use QoS network building blocks to provide
required end-end QoS
• Looking at:
– RSVP (Resource ReSeVation Protocol)
– RSVP over ATM
– IP Differential Services
– Winsock2 (support for native ATM and RSVP)
– Queuing technologies (Weighted Fair Queuing, etc)
– H.323 Gatekeeper
– Multimedia Conference Manager
– Sub-net Bandwidth Manager
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Performability Approach
• Characterise applications traffic / QoS features
– two-party and multi-party
– multicast & broadcast
– include end-system performance
• Measure operation of QoS techniques
– e.g. RSVP implementation in routers
• Match
– network performance required for user acceptance
– network performance achievable with QoS methods
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Example Performability Result
• Compare NetMeeting using Best Effort & RSVP
Radcom
Analyser
NetMeeting
Host #1
r s
NetMeeting
Host #2
r s
Serial Link
Investigate
Router
- BE (FIFO queue)
- Fair Queueing
- Reservation
- Controlled Load (Video)
- Guaranteed Service (Audio)
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Router
SmartBits Load Generator
Experiment & (Early) Results
• Examine end-end delay between hosts for audio
(G.723.1) & video (H.263) traffic
– 1) no background load on serial link
– 2) background load
, fair-queuing in router
– 3) background load
, RSVP reservations
(1250byte packets)
(1250byte packets)
• Audio delay
– 1) 1.5 ms to 28.4 ms, mean 8.8 ms
– 2) 1.5 ms to 39.5 ms, mean 9.0 ms
– 3) 90% between 1.2ms and 30ms
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Experiment & (Early) Results
• Video delay
– 1) mean 33.1 ms
– 2) 30 ms to 800 ms
• Not acceptable to users
– 3) 3.7 ms to 30ms
• Acceptable to users
• RSVP used to meet
users’ requirements
AIMS’99 Workshop
Heidelberg, 11-12 May 1999
Any Questions?
AIMS’99 Workshop
Heidelberg, 11-12 May 1999