Transcript Quality of Experience

End-to-end QoE Optimization Through
Overlay Network Deployment
Bart De Vleeschauwer, Filip De Turck,
Bart Dhoedt and Piet Demeester
Ghent University - Department of Information Technology
Maarten Wijnants and Wim Lamotte
Hasselt University - Expertise Centre for Digital Media
End-to-end QoE Optimization Through Overlay Network Deployment
Outline
• Introduction & Motivation
• Overlay Network Architecture
–
–
–
–
Architectural Overview
Overlay server
Overlay access component
Network Intelligence Proxy
• Evaluation
– Evaluation Testbed
– Experimental Results
• Conclusions
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
2
Introduction & Motivation
• Continuous rise in usage Internet for
multimedia services
– E.g. IPTV, Skype, online gaming
• Multimedia services are more sensitive to
packet loss, bandwidth restrictions, delay
and jitter
– Network performance guarantees required
• Failure to meet requirements will have
impact on user Quality of Experience
– E.g. jitter in VoIP = varying response times
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
3
Introduction & Motivation
• Internet is currently not able to provide
any performance guarantees
– Routing is best-effort
• Can result in suboptimal routing between hosts
• Solutions proposed to provide QoS (e.g. IntServ and
DiffServ) still suffer from unresolved issues
– Last mile can impose bandwidth bottleneck
• May result in congestion
• Increase in packet loss and packet delay possible
 No E2E solution for QoE provision yet
available
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
4
End-to-end Architectural
Overview
Overlay
layer
Network
layer
Last mile QoE optimization
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
5
Overlay Server (OS)
• Monitors connectivity through active
network probing
– Deduce delay & packet loss on overlay edges
• Maintains a virtual topology using overlay
routing tables (ORTs)
– ORTs map target OS to next hop OS
• Overlay routing tables are adapted to
reflect the current network condition
• Forwards overlay packets via “good” path
 OSs provide resilient overlay routing
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
6
Overlay Access Component
(AC)
• Deployed at the edge of the network
– End-device, residential gateway, …
• Monitors quality of connection (i.e. packet
loss & delay) between source-destination
• Packets are pushed to overlay layer in
case of problematic direct IP connection
– Packets pushed to closest OS
– OS forwards packets to AC close to target
– AC transfers packets back to IP network layer
 Transparent access to overlay network
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
7
Overlay Routing Example
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
8
Network Intelligence Proxy
(NIProxy)
• Optimizes last mile content delivery to
clients
– Through awareness introduction in network
• Both network- and application-related context
• Provides client bandwidth management
– Through creation stream hierarchy
• Acts as service provision platform
– Supports processing of multimedia flows
– Through generic plug-in approach
– E.g. video transcoding plug-in
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
9
Evaluation Testbed
• Setup with 10 Linux PCs
• Core network and last mile impairment
are emulated using Click nodes
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
10
Experimental Results –
Resilient Overlay Routing
• Packet loss ratio per
sec during 50 secs,
with and without
overlay routing
• 10% average loss
• OR: packet loss
circumvented
• SR: packet loss,
resulted in visual
artifacts at client-side
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
11
Experimental Results –
Last Mile QoE Optimization
• NIProxy managed last mile video delivery
• Experiment consisted of 5 intervals
– Caused by both bandwidth fluctuations and
client-initiated shifts in stream importance
No
NIProxy:
less optimal
e.g. video
hitches
Finding
2:
1: more
downstream
bw forQoE,
important
capacity
respected
streams
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
12
Conclusions
• Overlay network for E2E QoE management
– Components in network core and at edge
– Improves standard network routing service
– Last mile QoE optimization
• Results illustrate effectiveness of approach
– Complementarity constituting components
– Elimination/decrease packet loss using resilient
overlay routing (e.g. avoiding failing links)
– Last mile congestion avoidance through
intelligent bandwidth management and
multimedia service provision
End-to-end QoE Optimization Through Overlay Network Deployment
23/01/2008
ICOIN2008
13
Thank you for your attention!
Any questions?
End-to-end QoE Optimization Through Overlay Network Deployment