Institute of Computer Science
Chair of Communication Networks
Prof. Dr.-Ing. P. Tran-Gia
SDN-based Application-Aware
Networking on the Example of
YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn,
Thomas Zinner, Phuoc Tran-Gia
www3.informatik.uni-wuerzburg.de
Motivation
Application
Control Plane
Application
Application
…
Northbound-API
Network
Control Plane
Controller
Network
Control Plane
Southbound-API
Data Plane

Use application state information to optimize the user experience
and resource management
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
2
Scenario
Virtual Channels
ISP A
Access
Network
CDN
ISP B
Provider Edge
Customer Edge

Leveraging SDN for network resource
management
 Dynamic shift of application flows
between available channels to enhance
quality of critical applications
 Specific Example: Concurrent download
and YouTube flows
 Maintaining a good YouTube quality
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
3
Pre-Buffered Playtime as QoE-Metric with YoMo
pre-buffered playtime

YoMo (YouTube Monitor) estimates the pre-buffered playtime
 Used as:
 Control input parameter for the application-aware approach
 Quality indicator to evaluate the management algorithms
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
4
Reference Testbed

Controller running “switch” application
 only one link between the two OpenFlow switches usable
 Maximum throughput: 10 Mbit/s
Control
Channels
Up-Link
10 Mbit/s
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
5
Resource Allocation Methods
Required Information
Source of Information
Round robin
Number of links
OF controller topology
manager
Bandwidth-based
Used bandwidth of
flows
OpenFlow statistics
Deep Packet Inspection Type of
traffic/application
DPI software
Application awareness
Application
Application state
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
Increasing Complexity
Method
6
Experiment Procedure

0s: Begin of experiment; YouTube video is started

60s: Start of interfering traffic

Iperf-generated TCP flows are used as interfering traffic

60s+x: Interfering traffic flows start with a 1s interval

420s: End of experiment
t [s]
0
60 61 62 …
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
420
7
Reference Case with 5 TCP Flows

Maximum throughput of 10 Mbit/s
 The YouTube flows do not have sufficient bandwidth available
 Stalling can not be prevented
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
8
Bandwidth-Based Approach

Maximum throughput: 50 Mbit/s
 Using flow statistics to calculate throughput
 Algorithm:
 Select link based on throughput information
 Switch flows with the highest throughput to another link
Control
Channels
Up-Link
50 Mbit/s
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
9
Bandwidth-Based Allocation with 15 TCP Flows

Flows are distributed among the different links
 Influence of the interfering traffic is visible
 The scheduler can maintain the pre-buffered playtime
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
10
Bandwidth-Based Allocation with 20 TCP Flows

Flows are distributed among the different links
 Too many flows in the network
 The YouTube-video stalls
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
11
Application-Awareness Approach

Application notifies the controller about incoming application traffic
 Information about the YouTube-stream is obtained through YoMo
 Algorithm:
 Classification of the flows into different priority classes
 The highest flow priority on a link determines the allowed
number of flows
Control
Channels
Up-Link
50 Mbit/s
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
12
Application-Aware Allocation with 50 TCP-Flows

The maximum possible throughput is reached
 Critical threshold tC at 20 s pre-buffered playtime
 Regular threshold tR at 35 s pre-buffered playtime
 Shift of interfering flows, if the buffer falls below the threshold
 No stalling of the YouTube video
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
13
Efficiency of the Application-Awareness Approach

Without interfering traffic the typical YouTube behavior is observed
 With 25 TCP flows an influence is visible
 With 50 TCP flows an intervention of the scheduler is necessary
 For 75 and 100 interfering TCP flows a similar behavior is visible
 Pre-buffered playtime can be maintained
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
14
Conclusion

Dynamic shift of application flows between available channels to
enhance quality of critical applications

Implemented SDN network resource management for YouTube
using Northbound-API

Application-aware approach outperforms conventional
mechanisms in terms of QoE

Future Work:
 Trade-offs between multiple critical applications
 Other QoS management mechanisms, e.g. OF 1.3 flow meters
 More types and sources of application (state) information
 Identification of a suitable standard Northbound-API realization
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
15
Questions and Comments ?
SDN-based Application-Aware Networking on the Example of YouTube Video Streaming
Michael Jarschel, Florian Wamser, Thomas Höhn, Thomas Zinner, Phuoc Tran-Gia
16