14. ComNets-Workshop 2007
Smart Caching for
Supporting Video Streaming
in Heterogeneous
Wireless Networks
Stephan Göbbels
Chair of Communication Networks, Faculty 6,
RWTH Aachen University, Germany
14. ComNets-Workshop, Mobil- und Telekommunikation
March 30th, 2007, Aachen, Germany
Stephan Göbbels, ComNets, RWTH Aachen University
Overview
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
• Motivation and Introduction
• Smart Caching
• User Scenario
• Queuing Model
• Results
• Conclusion and Outlook
Stephan Göbbels, ComNets, RWTH Aachen University
2
Motivation
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
Data
• Mobilizing services and
applications:
Rate
Unused capacity
Wireless
– Video on Demand
– Online VCR
– IPTV
• Broadband wireless
coverage vs.
heterogeneity
• Discrepancy between
capabilities of fixed and
wireless networks
• High bandwidth
requirements –
low delay constraints
Backbone
t
Overload
 Adapt wireless networks
 Decouple wired and wireless
networks
Stephan Göbbels, ComNets, RWTH Aachen University
3
Introduction
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
•
Video streaming requires high
average data rates
– More than cellular networks can provide
– Smaller than WLAN peak data rates
•
Data rate varies due to
•
Life on buffered data in phases of no coverage
•
Legacy buffering in end device is not enough
•
Maximum network utilization is necessary
– Used wireless network
– Distance to Access Point (AP)
– Be greedy – Use always as much network resources as possible
– Requires user data provision
Stephan Göbbels, ComNets, RWTH Aachen University
4
Smart Caching
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
•
Buffering at the edge of the core network
•
Separation of end-to-end link
– One hop in the (wired) backbone
– Second hop (mostly) in the wireless network
•
On top of transport layer
– Packets leave transport layer, buffered and forwarded
•
Reuse of data requires clustering of access points
•
Protocol overhead kept away from backbone
•
Fast reaction on changing link
conditions
Server
Wireless Connection
Smart Cache
•
Integration of different networks
possible
•
In case of enough bandwidth no buffering necessary
Client
Backbone Connections
(wired)
Stephan Göbbels, ComNets, RWTH Aachen University
5
Application Scenario
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
• Applied service:
• Urban environment
Video streaming (varying
data rates)
• WiMAX radio network
• Varying coverage:
Server
• Cell radius: 50m
10-100%
• User speed:
IP world
1 m/s
Smart
Cache
AP4
AP1
AP3
Client
AP2
gap
Stephan Göbbels, ComNets, RWTH Aachen University
6
Queuing Model
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
• Emulation of service interrupt with “gap packets“
• Arrival process: Poisson and batch arrival (more appropriate for
MPEG streams)
• Translation of
residence probability
into path probability
p1;µ1
• µ given by different
PHY modes of
WiMAX
• Packet size:
100 Byte
λ
• Analytical model
allows the calculation
of Waiting Time
distribution
W
p2;µ2
p3;µ3
pgap
p1 µ
1
pWiFi
p2
µ2
p3
pgap
µ3
µgap
Stephan Göbbels, ComNets, RWTH Aachen University
7
Simulation Results
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
• Average Waiting Time
corresponds to the required
lead time for video
streaming
• Smart caching focuses on
delay insensitive traffic
1
Throughput
Throughput
Throughput
Throughput
Throughput
Throughput
Required Coverage [%]
0.9
0.8
=
=
=
=
=
=
21.8 Mbit/s
20.0 Mbit/s
15.0 Mbit/s
10.0 Mbit/s
5.0 Mbit/s
1.0 Mbit/s
0.7
0.6
• Less delay tolerant
applications require more
coverage
• Usual video streaming
rates allow small
coverage portions
• No loss of data
0.5
0.4
0.3
0.2
0
100
200
300
400
500
600
700
800
900
1000
• The closer the systems
comes to the maximum
throughput rate the more
ineffective Smart Caching
gets
Average Waiting Time [s]
Stephan Göbbels, ComNets, RWTH Aachen University
8
Simulation Results
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
1
• Strict limits require
almost full coverage
-> No Smart
Caching
• Lowest curve is limit
of available
throughput
0.9
Required Coverage [%]
• Waiting Time
boundaries imply
certain coverage
rates
0.8
0.7
0.6
0.5
0.4
Average W
Average W
Average W
Average W
Average W
0.3
0.2
0.1
0
5
10
15
=
=
=
=
=
1.0 s
10.0 s
100.0 s
1000.0 s
10000.0 s
20
25
Throughput [Mbit/s]
• ~22 Mbit/s is maximum
data rate of the scenario
Stephan Göbbels, ComNets, RWTH Aachen University
9
Simulation Results
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
• Which influence has the
traffic modeling? Compare:
• Smart Caching ->
Smoothing of arrival
process
– Poisson stream
– MPEG stream (batch
arrival)
15
• Change in average waiting
time
• Effect increase with raising
data rate
• For coverage portions of
less than 70% Poisson
arrival is sufficient
Relative Impact [%]
• Less influence of
interarrival time variance
Throughput: 1 Mbit/s
Throughput: 2 Mbit/s
10
5
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Network Coverage [%]
Stephan Göbbels, ComNets, RWTH Aachen University
10
1
Conclusion and Outlook
Overview – Introduction – Smart Caching – Scenario – Queuing Model – Results – Outlook
• Smart Caching improves performance of
heterogeneous networks
– Patchy network coverage
– Variance in link quality
• It is suitable for video streaming services:
– Video on Demand (VoD)
– IPTV
– E.g.: Bandwidth consuming delay tolerant traffic
• No full network coverage is required
• Future:
– Restrict transfer of streaming data to high performance
areas
– Scheduling based on delay requirements and available
bandwidth
Stephan Göbbels, ComNets, RWTH Aachen University
11
Thank you for your attention !
Stephan Göbbels
[email protected]
Any questions?
Stephan Göbbels, ComNets, RWTH Aachen University
12