Enhancing Quality of Service
in Software-Defined Networks
Francesco Ongaro
ALMA MATER STUDIORUM - UNIVERSITY OF BOLOGNA
Department of Computer Science and Engineering
Master Degree in Computer Engineering
Supervisor: Professor Antonio CorradiCorrelator: Professor Mario GerlaCorrelator: Professor Eduardo CerqueiraCorrelator: Doctor Luca Foschini
Francesco Ongaro
2Outline
1. Introduction
2.Background
3.QoS-aware Model
4. Implementation
5.Results
6.Conclusion
Francesco Ongaro
3Introduction (I)Motivations
Internet ossification phenomenon: while technology and innovation continue to evolve, our network infrastructure system has been maintained almost in the same shape for decades.
Quality of Service (QoS) guarantees: the explosive growth of real time applications brings the network programmers to design network protocols that deliver certain QoS performance guarantees.
Coexistence of differentiating services with different requirements in terms of bandwidth, packet loss, latency, and also various implementation technologies in campus networks and every day scenarios.
Francesco Ongaro
4Introduction (II)Quality of Service
T. Szigeti and C. Hattingh, End-to-End QoS Network Design: Quality of Service in LANs, WANs, and VPNs (Networking Technology). Cisco Press, 2004.
APPLICATION DELAY PACKET LOSS JITTER BANDWIDTH
VoIP ≤ 150 ms * ≤ 1 % < 30 ms * 21-320 kbps
Interactive-Video ≤ 150 ms * ≤ 1 % ≤ 30 ms * n.a.
Streaming-Video ≤ 4-5 s * ≤ 5 % n.a. n.a.
* "one-way" value
Francesco Ongaro
5Introduction (III)Software-Defined Networking paradigm
SDN together with OpenFlow allows to control the behavior of the network by decoupling the control logic and the physical layer.
Francesco Ongaro
6Background (I.I)“Classical” Switch
Francesco Ongaro
7Background (I.II)SDN Switch
Francesco Ongaro
8Background (II)OpenFlow-enabled Switch
pack
et-i
n flow
en try
Francesco Ongaro
9Our QoS Architecture
Francesco Ongaro
10QoS aware ModelMulti-Commodity Flow and Constrained Shortest Path
Objective Function:The cost minimization that depends on the cost computed with the delay and packet loss of the used links.
Constraints: The flow balancing, that the total flow incoming into each vertex is equal to the total flow outgoing from the same vertex, with the exception of the source and the terminal.
Francesco Ongaro
11QoS aware ModelMulti-Commodity Flow and Constrained Shortest PathConstraints: The arch capacity constraint imposes a limit on the available bandwidth of each link.
The maximum acceptable value for the packet loss Pkmax and the
delay Dkmax, imposes the limit for each service k.
The variables domain guarantees that the decision variable is 0 or 1.
Francesco Ongaro
12Implementation (I)Scenario at UCLA campus
Two different flows: medical data, e.g., file transfer general data, e.g., multimedia service
Various paths with different characteristics related to the packet loss, delay, and available bandwidth.
In case of network congestion, it is necessary to: Reconfigure the paths. Guarantee the route for some specific flows instead of other.
[UCLA campus map]
Francesco Ongaro
13Implementation (II)Hybrid Network Topology in Mininet
Francesco Ongaro
14Implementation (III)Details
Francesco Ongaro
15Results (I)Short and Permanent Congestion
Time (s)
Time (s)
Francesco Ongaro
16Results (II)Triggering
The Watch Dog module triggers the changing of the path when the throughput is under the chosen threshold.
We call these specific warnings: “false-positive”.
Time (s)
Francesco Ongaro
17Results (III)QoS aware (i)
Time (s)
Link congestion during the video streaming service.
New path followed by the video streaming flow to avoid the link congestion.
Francesco Ongaro
18Results (III)QoS aware (ii)
Time (s)
New link congestion during the video streaming service.
Second new path followed by the video streaming flow to avoid the new link congestion.
Francesco Ongaro
19Results (IV)Multi-Commodity Flow with QoS
Link congestion during both the video streaming and the file transfer.
Time (s)
New path followed by the video streaming and file transfer flows to avoid the link congestion.
Francesco Ongaro
20Results (V)Mean Opinion Score (MOS) and Function Cost
FUNCTION COST MOS QUALITY IMPAIRMENT
< 40 5 Excellent No block at all
40 – 55 4 Good No block or a sporadic very short block
56 – 69 3 Fair A couple of short blocks (1 - 2 s)
70 – 79 2 Poor Several blocks (2 - 4 seconds long)
> 80 1 Bad A lot of blocks (> 10) with long duration (7 - 10 s)
Francesco Ongaro
21Conclusion We develop a modular and extensible architecture to enhance the Quality of
Service in a Software Defined Networks.
We allow dynamic and fast path reconfiguration in case of network overloading taking into account the service requirements such as packet loss, delay, and available bandwidth.
We map the results given by our mathematical model into a MOS scale in order to represent the client perception of the service.
As Future Work:
We plan to use our QoS architecture to manage a seamless vertical handover between different technologies such as Wi-Fi, WiMAX, and LTE.
We think that our QoS Architecture can improve the resource management and the path reconfigurability in a very dynamic scenario such as the Vehicular Ad-Hoc Networks (VANETs)*.
* I. Ku, Y. Lu, E. Cerqueira, R. Gomes, F. Ongaro, and M. Gerla, “Towards Software-Defined VANET: Architectures and Services,” Med-Hoc-Net 2014.
Top Related