Project topics and materials
Get Complete Project Material File(s) Now! »
Basic Characteristics of the Residential Traffic
To give an overview of the data, we start by showing the evolution of volume and number of users over the week, we plot each days on Fig. 3.1. The curve of the volume (Fig. 3.1a) shows a very similar pattern over the week. Thursday the 7th has the highest volume whereas Sunday the 9th has the lowest. As for the number of users (Fig. 3.1b), there are some differences between the days: for example, Saturday night has the least amount of users.
In Fig. 3.2, we trace the evolution of volume (TCP) over two different days of the week. On a week day (Fig. 3.2a), we observe an unsurprising camel curve with 2 large peaks around the mid-day break. Whereas on Sunday (Fig. 3.2b), we have a large plateau from late morning to early afternoon.
DailyMotion Mean Flow Rates
As for DailyMotion in Fig. 4.9a, we have very homogeneous mean rates in all traces that show a large accumulation point just above the median video encoding rate at 500 kb/s, except for 2008/07 trace. Thus, even if there is no peak rate limitation (see Fig. 4.6a), there is a mean rate limit for DailyMotion videos set slightly above the median video encoding rate. While such a choice of the rate limit should allow for a correct reception (and viewing) quality for most of the videos, the reception can be very sensitive to any network problem that may cause the reception rate to fall below the encoding rate for some limited time. In the FTTH M trace of 2008, we see that the mean rate limit originally was higher at about 12Mb/s.
Such modifications in the rate limitation policies made by the video sharing sites are usually not known in advance to the ISP. However, they may have an important impact on the network: unlimited peak rates and moderate mean rates may lead to much more bursty traffic arriving in the router queues. We would like to emphasize the fact that the peak rate and mean rate limits of TCP flows are independent: indeed, DailyMotion has no peak rate limit (at least up to 100Mb/s) but a strict mean rate limit at 500kb/s.
YouTube Mean Flow Rates
In Fig. 4.9b, we can see that the policy concerning the mean rate limitation of You- Tube has evolved over time. For the 2008/07 trace, there is a sharp mean rate limit at 1.2Mb/s that has been previously observed [41]. Such a limitation of both peak rate and mean rate, as in the case of YouTube, was most likely implemented using a well-known open-source rate limiter, the Token Bucket Filter over Hierarchical Token Bucket (HTB [29]) with two buckets (one limiting peak rate and one limiting mean rate). Note that YouTube uses a new distribution policy since 09/2010, so the conclusions on YouTube peak and mean rates do not hold any more. The FTTH M 2009/12 afternoon trace achieves average flow bit-rates superior to the median video encoding rate for 95% of the videos. As for mean rate, the shape of the graphs does not allow to infer any mean rate limitation. For the traces taken in the evening, around 40% of the videos achieve a mean reception rate that is inferior to the median encoding rate. The curves are concave with no clear limitation. As we will see later (in Tab. 4.5), such low reception rates result in bad reception quality.
In Fig. 4.10, we closer
Table of contents :
List of Figures
List of Tables
1 Introduction
1.1 Network Measurement: Tomography
1.2 ISP Motivation
1.3 Organisation of the Thesis
I Passive Measurements
2 Passive Measurements Context and Methods
2.1 Methods and Tools for Passive Measurements
2.2 Contributions
2.2.1 Analysis of one week of ADSL connections
2.2.2 HTTP Video Streaming Performance
2.3 Related Work
2.3.1 Network Tomography
2.3.2 Video Streaming Studies
3 Analysis of one Week of ADSL Traffic
3.1 Data Collection
3.2 Characteristics of the Residential Traffic
3.2.1 Application Share
3.2.2 Refined Application Distribution
3.2.3 Streaming Analysis
3.2.4 Facebook
3.2.5 YouTube
3.2.6 Volumes
3.2.7 Users’ Sessions
3.2.8 User’s Level Analysis
3.2.9 Performance Analysis
3.3 Clustering Analysis
3.3.1 “Real” vs. “fake” usage
3.3.2 Choice of clustering
3.3.3 Impact of Timescale on the Clustering Analysis
3.3.4 Conclusion of the Users Clustering Analysis
3.4 Dimensioning
3.4.1 P2P Rate Limit at Peak Hour
3.4.2 P2P Rate Limit at Peak Hour for Heavy Hitters only
3.4.3 Conclusion on Dimensioning
4 HTTP Video Streaming Performance
4.1 Novelty of this Work
4.2 Trace Characteristics
4.3 HTTP Streaming Context
4.3.1 Most Popular Video Sharing Sites
4.3.2 Video Encoding Rate
4.3.3 Domain Name System (DNS)
4.3.4 Distribution of Traffic across ASes
4.4 Flow Performance Indicators
4.4.1 Round Trip Time
4.4.2 Peak Rate
4.4.3 Mean Flow Rates
4.4.4 Loss Rate
4.4.5 Methodology for Monitoring
4.5 User Behavior Study
4.5.1 Downloaded Duration
4.5.2 Simple User Experience Metric
4.5.3 How do Users watch Videos
5 Conclusion of Part
5.1 Conclusions on the Analysis of Week-long Connection Statistics
5.2 Conclusions on the Performance of HTTP Video Streaming
5.2.1 YouTube Architecture and Video Servers Selection
5.2.2 DailyMotion Delivery Policy
5.2.3 Users’ Viewing Behavior
5.2.4 Next Steps on Utilising Passive Packet Traces to Understand
Video Streaming Traffic
II Active Measurements
6 Active Measurements Context and Challenges
6.1 Active Measurements of HTTP Video Streaming
6.2 Related Work
6.3 Contributions
6.3.1 Main Results
6.3.2 Novelty of our Work
7 Impact of YouTube Delivery Policies on the User Experience
7.1 Methodology
7.1.1 Tool Presentation
7.1.2 Validation Process
7.2 Datasets Details
7.2.1 Volunteer Crawls
7.2.2 Controlled Crawls
7.2.3 Kansas City Crawls
7.3 Results
7.3.1 Video Server Selection
7.3.2 DNS impact
7.3.3 Evaluation of QoE Approximation Techniques
7.4 YouTube Infrastructure
7.4.1 Datacenters sizes
7.4.2 Redirections
8 Conclusion of Part II
9 Conclusion
III French Summary
10 Introduction
10.1 Mesure du réseau Internet
10.2 Le point de vue des opérateurs
10.3 Organisation de la thèse
11 Principales Contributions
11.1 Mesures Passives
11.1.1 Analyse d’une semaine de trafic ADSL
11.1.2 Analyse de la performance du vidéo Streaming
11.2 Mesures Actives
11.2.1 Outil d’évaluation de la qualité d’expérience
11.2.2 Présentation des données collectées
11.2.3 Résultats
12 Conclusion
Bibliography
