Internet Streaming and Network Neutrality:

Comparing the Performance of Video Hosting Services

Alessio Botta

1,2

, Aniello Avallone

, Mauro Garofalo

and Giorgio Ventre

University of Napoli Federico II, Napoli, Italy

NM2 srl, Napoli, Italy

Keywords:

Network Neutrality, Internet Video Streaming, Network Performance.

Abstract:

Network neutrality is a hot topic since a few years and involves different aspects of interest (e.g. economic,

regulatory and privacy) for a wide range of stakeholders, including policy makers, researchers, economists,

and service providers. When referring to video streaming, a killer web service of the Internet, much has been

discussed regarding if and how video providers violate or may violate neutrality principles, in order to give

users a “better” service compared to other services or to other providers. In this paper we provide a contribution

to this discussion analyzing the performance of three main video hosting providers (i.e. YouTube, Vimeo, and

Dailymotion) from an user viewpoint. We measure the throughput and RTT experienced by users watching

real videos of different popularity, at different day hours and at several locations from around the world. We

uncover the performance differences of these providers as a function of the different variables under control

and move a step forward to understand what causes such differences. Our results allow to understand what are

the real performance users currently get from these providers and if the performance differences observed can

be due or to considered as a violation of network neutrality principles, providing a ground for people interested

in legal and regulatory issues of web applications and services.

1 INTRODUCTION

There is a long ongoing debate on network neutral-

ity, for which several deﬁnitions exist which share

the common idea that data on the Internet should be

treated in the same way despite several its character-

istics such as technology, device, application, service,

user, provider, and country they come from or go to.

A ﬁrst debate about network neutrality in terms of In-

ternet trafﬁc management policies appeared in 2003

(Wu, 2003), but concerns about possible threats to the

end-to-end nature of the Internet raised already in the

late 1990s (Lemley and Lessig, 2000). Nowadays the

debate has gained momentum also because of recent

events such as the one involving the provider Com-

cast, which was slowing uploads from peer-to-peer

ﬁle sharing applications (Svensson, 2007). The dis-

cussion on whether the Internet should be fully neu-

tral, or rather the providers should be allowed to use

techniques to differentiate the trafﬁc don’t concerns

just economic but increasingly both legal and regula-

tory aspects. A work regarding law aspects was pre-

sented in (Koops and Sluijs). In this paper we do

not want to take a position pro or against network

neutrality. We rather aim at providing a contribu-

tion to understand the current situation from a user

viewpoint, which is of interest for people concerned

with legal and regulatory issues of web applications

and services. Our work focuses on three Video Host-

ing Services, YouTube, Vimeo and Dailymotion, for

which we studied the performance achievable by end

users depending on video popularity and user loca-

tion (i.e. country). The highlights of our work could

be summarized as follow:

• We introduce a methodology that, regardless of

the providers considered in this work, allows ac-

quiring and analyzing performance statistics.

• We measure, analyze, and compare statistics pro-

vided by YouTube, Vimeo, and Dailymotion,

from several locations all around the world.

• We provide insights on geographical location of

the infrastructures and routing policies used by the

video hosting services to deliver their content.

The reminder of the paper is structured as follows.

Related Work is reviewed in the next section, which

also highlights the novel aspects of this work. Knowl-

edge on infrastructure of video hosting services is

514

Botta, A., Avallone, A., Garofalo, M. and ventre, G.

Internet Streaming and Network Neutrality: Comparing the Performance of Video Hosting Services.

DOI: 10.5220/0005798705140521

In Proceedings of the 2nd International Conference on Information Systems Security and Privacy (ICISSP 2016), pages 514-521

ISBN: 978-989-758-167-0

given in Section 3. In Section 4, we present our

methodology, and how we collected the dataset. Sec-

tion 5 describes in detail the results and it is followed

by a discussing about geographical location in Section

6. We conclude the paper in Section 7.

2 RELATED WORK

Several interesting works on analysis of video host-

ing services are focused on YouTube only, since it

generates a large share of Internet trafﬁc. The ﬁrst

extensive data-driven analysis about video popular-

ity and users behavior of YouTube was presented in

2007, (Cha et al., 2007). Data collection involves a

very long time period (tens of years) and furthermore

compares YouTube with classic on demand video

providers such as Netﬂix and Yahoo! Movies. A tool

to measure QoS and QoE of YouTube is designed in

2012, (Plissonneau et al., 2012). Metrics collected by

a hundred of volunteers, have been analyzed by au-

thors to infer the main delivery policies of YouTube

videos, to understand the impact of the ISP on these

policies, and ﬁnally YouTube policies are compared

in the US and Europe. One of the earliest analysis

of HTTP video streaming with a comparison between

YouTube and one of its competitors Dailymotion was

presented in 2012, (Plissonneau and Biersack, 2012).

They use packet traces from a residential ISP network

to infer for each streaming ﬂow the video character-

istics, such as duration and encoding rate, as well as

TCP ﬂow characteristics, such as RTT and packet loss

rate. More focused on geographical location of in-

frastructure, (Padmanabhan and Subramanian, 2001)

has built a service to translate the IP addresses of In-

ternet hosts into their geographical location. They

have proposed three techniques to infer the location

of target host. GeoTrack, based on information pro-

vided by DNS. GeoPing, using delay measurements

between target host and geographically known loca-

tion, and GeoCluster, that combines partial host-to-

location mapping information and BGP preﬁx. A re-

cent work, (Calder et al., 2013), try to clustering all

servers of Google infrastructure in serving-site and

then localize them using a technique called Client-

Centric Geolocation or CCG. The CCG is based on

the hypothesis that clients that are directed to the

server are likely to be topologically, and probably ge-

ographically, close to the server.

Summarizing, studies more relevant to our work in-

vestigated either the CDN infrastructure and perfor-

mance measures, or the geographic location of such

infrastructure. Our work moves a step forward with

respect to existing literature. To the best of our knowl-

edge, we are the ﬁrst to provide a comparative anal-

ysis of the three most popular video hosting services,

YouTube, Vimeo, and Dailymotion. We uncover the

different performance they provide to real users from

around the world and investigate on the causes of such

differences, providing insights on the infrastructure

each of them uses for video delivery. Unlike works

based on residential ISPs measurements, that involve

a large number of volunteers, we perform active mea-

surements using a globally distributed research infras-

tructure (i.e. (Chun et al., 2003)). Performance indi-

cators collected may be different from the ones of res-

idential users. However, our main aim of comparing

the different services (to understand if and how they

violate or may violate network neutrality principles)

is not affected by this choice.

3 INFRASTRUCTURES OF

PROVIDERS

The following section describes the infrastructure of

the video hosting services analyzed. Although there

are several studies focused on YouTube (Google) in-

frastructure (e.g. (Calder et al., 2013)), that try to

identify the number, structure, and location of caches

and servers, less information is available on the other

providers. Usually, the information about these in-

frastructures is not publicly disclosed. We crossed

several sources of information for obtaining the views

on the infrastructures that follow. Then we verify and

conﬁrm the accuracy of such views with the experi-

ments in Section 6.

3.1 Dailymotion

Dailymotion was launched in France in 2005. Orig-

inally it consist of one homemade Linux cluster and

limited connectivity via a classical Internet connec-

tion able to serve only a few thousand users (Pelaprat,

2007), (EMC2, 2010). Afterwards it moved to a

more scalable architecture for storage and simultane-

ously read/write on a network ﬁle system where In-

put/Output bandwidth, caching and latency are shared

throughout the system and performance scales lin-

early with the numbers of nodes. Dailymotion has

not entered into agreements with any third party CDN

but, starting 2014, it has chosen Orange as partner for

the optimization of worldwide distribution of video

content, as part of its launch of premium live stream-

ing “channels” (Orange, 2014). A solution, called

“Media Delivery”, that uses an extensive network of

servers to accelerate the distribution of video content

over the Internet, was implemented by Orange and

Internet Streaming and Network Neutrality: Comparing the Performance of Video Hosting Services

515

Akamai Technologies(Akamai, 2014a). At present

Dailymotion provides an “only for premium services

platform” while standard users are connected to the

origin data center, with a data discrimination based

on content.

3.2 Vimeo

Vimeo, founded in 2004, to distribute its content uses

the Akamai CDN (Vimeo, 2013). Akamai had a

broadly deployed network of edge servers, with 20

to over 100 times more Points of Presence (POPs)

than other global CDN providers (Akamai, 2014c).

Its edge servers are located deep within thousands

of ISPs networks, as close as possible to the users,

through the partnerships with the leading Internet ser-

vice providers. They claim to providing lowest la-

tency, high throughput and low risk of network con-

gestion. Akamai provide a CDN dedicated to the

streaming media content deployment, named Adap-

tive Media Delivery (Akamai, 2014a). It allows the

transmission of video streams with Adaptive Bit Rate

and back up it in Akamai NetStorage, for later view-

ing (Akamai, 2014b). Akamai has also developed a

modiﬁed version of TCP/IP to optimize the transmis-

sion speed. This protocol called Fast TCP uses the

delay as a measure to control network congestion and

improve the throughput. The beneﬁcial use of this

protocol are exploited by the CDN in the accelera-

tion of both video distribution and download (Aka-

mai, 2012).

3.3 YouTube

YouTube born in 2005, was bought by Google in

2006. It is the most popular service on which users

can share and watch video content. The infrastructure

can be organized in the following components:

• Data Center: a set of high-efﬁciency Backend

servers used for computation and storage.

• Edge Points of Presence (POPs): cache servers

distributed worldwide,(Google, 2015). PoPs rep-

resent the terminal nodes of Google network

and are connected via peering with ISPs to

deliver Google services trafﬁc to users. The

caches are identiﬁed by Google in four logi-

cal namespace and PoPs are classiﬁed based

on a three level hierarchy. Primary cache

cluster: a

c.v[1-24].lscache[1-8].c.youtube.com,

where “a c” matches the IATA airport code

(IATA, 2010). Secondary cache cluster: tc.v[1-

24].cache[1-8].c.youtube.com. Tertiary cache

cluster few number of cluster named as cache or

altcache.

• Backbone: a global ﬁber network to interconnect

data centers and deliver trafﬁc to Edge PoPs.

• Google’s edge caching the whole cache infras-

tructure including nodes inside the ISPs. These

nodes (PoP) are calling Google Global Cache

(GGC) (Calder et al., 2013) and allow ISPs to de-

liver Google contents to the users, increasing per-

formance and reducing transportation costs being

more close to each other.

4 METHODOLOGY AND TOOLS

Scope of our work is to evaluate the performance of

video hosting services to understand whether such

performance differences could impact on network

neutrality and users privacy. We evaluate the perfor-

mance of video hosting service provider for different

kinds of content and different geographical location

to discover if: the providers have their own infrastruc-

ture in the country; there are cache-servers deployed

inside ISP infrastructures; special routing policies ex-

ist and how/when they are applied. The web service

under test are YouTube, Vimeo, and Dailymotion. We

analyze the trafﬁc related to video downloads because

streaming video accounts the majority (about 60%) of

Internet trafﬁc (Cisco, 2015). We have deﬁned four

categories of videos depending on popularity: less

than 500 views, between 10k and 120k views, be-

tween 120k and 1M views, and over a million views.

We have chosen one video with a 720p resolution

for each category. Every experimental campaign is

performed over a period of one day (24 hours) and

downloads are carried out at intervals of two hours.

The data acquisition phase of our analysis lasted 15 of

weeks in a period of time that covers several months.

Data reported in the following section refers to one

of these campaigns. Results of the other campaigns

showed similar results. As set of geographically well

know clients we used a distributed network of 200

PlanetLab nodes, deployed on a total of 36 countries.

PlanetLab uses high speed networks inside Research

Centers and Universities therefore the analysis cannot

strictly describe behavior of residential users. How-

ever, our aim is to compare the performance of dif-

ferent providers and we use PlanetLab as a reference.

More in depth, for each PlanetLab node we perform

the following batch operations to acquire the data:

• Running netstat in background. This command

is used to detect the IPv4 address of server that

physically contain the video to which the client

has been addressed by DNS.

• Using youtube-dl to download the video. This

ICISSP 2016 - 2nd International Conference on Information Systems Security and Privacy

516

command is used to estimates the throughput.

• Given the IPv4 address of server that physically

contains the video, running a ping to evaluate RTT

and TTL.

• Using traceroute tool to discover the path from

client to server and mapping the name of routers

in the path.

We assume that the measures of RTT and TTL,

as well as the path shown by traceroute, could be

in a good approximation representative of the net-

work status during the video downloads. Video

providers can adopt various approaches to cope the

fragmentation of terminals and network connection

issues. Adaptive bitrate streaming (ABS) is one of

the widespread technique that adapt the bitrate in re-

sponse to changing bandwidth conditions. Among

the providers under test, only Vimeo does not uses

a web player that supports ABS for video delivery.

So, even if youtube-dl support DASH (Lederer et al.,

2012) and HLS (Pantos and May, 2015), two spread

ABS standard implementation, we have performed

video download without taking advantage of these

new techniques in order to compare the providers at

conditions that are as similar as possible to each other.

5 EVALUATION

In the following section we analyze three video host-

ing services, Dailymotion, Vimeo, and YouTube,

evaluating performance indicators (i.e. throughput,

RTT, and distance in hops) related to users download-

ing videos of different popularity from several loca-

tions all around the world.

5000

10000

15000

20000

25000

30000

35000

40000

dm-1M

dm-120K

dm-10K

dm-500

vi-1M

vi-120K

vi-10K

vi-500

yt-1M

yt-120K

yt-10K

yt-500

KiB/s

Stat

Median

Mean

Figure 1: Throughput - Comparison among Dailymotion,

Vimeo and YouTube.

-200

-100

100

200

300

400

500

600

dm-1M

dm-120K

dm-10K

dm-500

KiB/s

Stat

Median

Mean

Figure 2: Throughput - Dailymotion Zoom.

Table 1: Throughput (KiB/s) - Statistical Indicators.

Provider-Class Min I Quart. Median Average III Quart. Max

dm-1M 21.36 353.8 415.1 487.1 502.4 11570

dm-120K 21.18 351.9 401.9 467.4 475.0 12690

dm-10K 24.10 344.1 378.5 421.8 419.9 8354

dm-500 6.752 354.6 402.7 474.2 471.6 9368

vi-1M 24.44 1788 5367 6944 9802 35010

vi-120K 18.63 1770 4992 6619 9000 34790

vi-10K 17.45 1976 6110 7806 10310 39000

vi-500 29.05 2144 6226 7894 10300 38830

yt-1M 11.79 1228 4408 5698 8235 23300

yt-120K 33.83 1153 4156 5930 8465 34690

yt-10K 32.3 1124 4014 5801 8617 34900

yt-500 14.06 1565 5025 6793 9034 31760

5.0.1 Throughput

Figure 1 depicts the values of throughput recorded

by clients for each provider and each video cate-

gory. Dailymotion (dm-*) has the lowest values

while Vimeo (vi-*) and YouTube (yt-*) have similar

throughput. The statistical indicators of throughput

(i.e. minimum, ﬁrst quartile, median, average, third

quartile, and maximum) refer to all providers are are

shown in detail in Table 1. Dailymotion has a median

value around 400 KiB/s while most of clients do not

exceed 500 KiB/s. Spikes affect the throughput av-

erage and these values are related to “anomaly” that

we will show more in depth in the following. Vimeo

and YouTube have similar trends although Vimeo has

higher throughputfor each video category as shownin

table 1. Also in these cases, the averages are affected

by spikes which signiﬁcantly differ from the median.

Regarding Vimeo, throughput of 320 Mbps and RTT

of 0.140 ms are shown. These spikes are related to

PlanetLab nodes who are only one “hop” away from

Vimeo servers. A brief discussion about these values

will be provided in the following.

5.0.2 Round Trip Time

The performance of providers concerning the delay

Internet Streaming and Network Neutrality: Comparing the Performance of Video Hosting Services

517

200

400

600

800

1000

1200

1400

1600

dm-1M

dm-120K

dm-10K

dm-500

vi-1M

vi-120K

vi-10K

vi-500

yt-1M

yt-120K

yt-10K

yt-500

Stat

Median

Mean

Figure 3: RTT.

(i.e. RTT) are shown in Figure 3. RTT values regard-

ing Dailymotion varying from 40 ms to 150 ms with

a median value of 100 ms. These values are always

higher in comparison to its competitors. This could

be related to the centralized nature of Dailymotion

infrastructure, Section 3.1. Regarding Vimeo, there

is a small range of variation, from 2 ms to 18 ms,

and the minimum values are very low independently

of video categories. An extremely low RTT value,

about 0.14 ms, has been recorded by PlanetLab node

planetlab1.arizona-gigapop.net,located in the United

States. A distributed infrastructure, like Akamai CDN

Section 3, allows to obtain better performance both in

terms of Throughput and RTT, regardless the country

from which the client requests the content. YouTube

have an analogue behavior, whose servers show RTT

values similar to those of Vimeo although on a wider

range (from 2 ms to 30 ms).

Table 2: RTT(ms) - Statistical Indicators.

Provider-Class Min I Quart. Median Average III Quart. Max

dm-1M 3.375 44.29 103.1 117.4 152.4 1013.0

dm-120K 3.568 44.95 103.7 116.2 149.8 1547.0

dm-10K 3.516 43.12 102.4 114.6 150.8 697.5

dm-500 3.42 45.35 102.1 116.0 150.9 473.4

vi-1M 0.135 2.378 6.738 16.91 16.72 421.1

vi-120K 0.141 2.500 7.183 17.45 18.33 405.1

vi-10K 0.139 1.887 5.698 14.28 15.21 473.7

vi-500 0.142 1.914 5.839 14.54 15.00 358.9

yt-1M 0.223 2.550 7.312 31.17 23.53 510.8

yt-120K 0.208 2.869 7.623 32.08 27.90 492.3

yt-10K 0.241 3.214 7.751 32.09 29.59 420.9

yt-500 0.205 2.405 6.935 28.92 19.58 488.5

5.1 Temporal Behaviour

In the following sections, we will point out how per-

formance evolve in time. This kind of analysis is

aimed at identifying differences of performance or

treatment for the different hours of day.

Figure 4 shows a whole day comparison between

the throughput of providers, related to the video cat-

egories. Regarding Dailymotion, Figure 4(a), there

are no particular treatments related to the different

video categories. Moreover the average of throughput

evolves in a nearly constant way. Standard deviation

value highlights spikes at 8h and 12h. They refer to

servers deployed in Korea and Singapore. They let

us suppose the presence of servers in these countries,

contrary to the assumptions made in the in Section

3. A further investigation to understand the perfor-

mance of PlanetLab nodes in these countries is left as

future work. The overlapping of the temporal evolu-

tion of the mean values and the standard deviations

of throughput is shown in Figure 4(b) for Vimeo and

in Figure 4(c) for YouTube. The median values are

constantly above 5000 KB/s, an order of magnitude

higher than Dailymotion. Moreover, all providers

have better performance for videos with lower num-

ber of views.

5.2 Performance by Country

This section will be shown the values of the perfor-

mance parameters according to the country of the

client. Data acquisition is performed over a period of

one day (24 hours) with tests carried out at intervals

of two hours.

Figure 5(a) shows the average throughput of each

video category in each country for Dailymotion.

The Figure depicts facts that support our hypothesis

about the centralized location of the all Dailymotion

servers: the performance is better in France than in

other countries; all European countries have higher

performance than non-European ones. We excluded

the values of Singapore and Korea (for Singapore the

maximum throughput is about 12000 KiB/s) other-

wise they would have made the graph unreadable. For

Vimeo, as we can see in Figure 5(b), the through-

put is variable in each country. Unlike Dailymotion,

these differences are not related to the distance be-

tween client and server, but to the quality of the net-

work in each country. Regarding the analysis of dif-

ferent video category, there is an overlap of the val-

ues of the average throughputin almost every country,

with higher values for the video below the 500 views,

denoting the presence of different treatments accord-

ing to the video category. Finally, Figure 5(c) shown

mean values and standard deviations of the through-

put of YouTube servers. Whereas the Google distri-

bution network is based on the concept of peering,

see Section 3, performance is strongly inﬂuenced by

the network infrastructure of the certain country. The

management through peering has led the CDN to de-

ﬁne supply agreements with third-party companies, in

order to obtain the widest possible capillarity. The

ICISSP 2016 - 2nd International Conference on Information Systems Security and Privacy

518

-400

-200

200

400

600

800

1000

1200

1400

0 5 10 15 20

KiB/s

Video 1M

Video 120K

Video 10K

Video 500

(a) Dailymotion

2000

4000

6000

8000

10000

12000

14000

16000

18000

0 5 10 15 20

KiB/s

Video 1M

Video 120K

Video 10K

Video 500

(b) Vimeo

-2000

2000

4000

6000

8000

10000

12000

14000

0 5 10 15 20

KiB/s

Video 1M

Video 120K

Video 10K

Video 500

Figure 4: Average Throughput over 24h.

200

400

600

800

1000

1200

1400

1600

1800

KiB/s

Video 1M

Video 120K

Video 10K

Video 500

(a) Dailymotion

-5000

5000

10000

15000

20000

25000

KiB/s

Video 1M

Video 120K

Video 10K

Video 500

(b) Vimeo

-5000

5000

10000

15000

20000

25000

KiB/s

Video 1M

Video 120K

Video 10K

Video 500

Figure 5: Average throughput in each country.

100

150

200

250

400

Video 1M

Video 120K

Video 10K

Video 500

(a) Dailymotion

-20

100

120

Video 1M

Video 120K

Video 10K

Video 500

(b) Vimeo

100

150

200

250

country

Video 1M

Video 120K

Video 10K

Video 500

Figure 6: Average RTT in each country.

comparison of mean values denotes variability related

to the country where surveys are carried out. As with

other providers, video with fewer views have perfor-

mance slightly better than others categories.

5.2.1 Round Trip Time

The average values and standard deviations of RTT,

regarding Dailymotion, are shown in Figure 6(a).The

overlap of the average values of RTT means that no

treatments to different category of video is applied.

The European countries have lower RTT than others

countries; in particular the average value is in most

cases below 50 ms while extra-European countries

have values always higher than 100 ms. Speciﬁcally,

Korea and Singapore RTTs are respectively 300 ms

and 350 ms. Using traceroute, we made an analysis

of the paths taken by packets traveling from the client

in Korea towards the provider’s server. First, pack-

ets pass by internal hops inside the network of the

Kookmin University (where are deployed the Planet-

Lab node), then they travel towards Lever 3 US and

Level 3 Paris and ﬁnally to the Dailymotion server.

The same analysis was repeated for the client in Sin-

gapore and results would conﬁrm for Dailymotion the

hypothesis of a centralized infrastructure, Section 3.

Concerning the RTT value of Vimeo, shown in Fig-

ure 6(b), in almost all countries the values are similar

for each video category of video. We can therefore

assume that every client is routed to the same servers

that contains all videos. The RTT for the majority of

countries is smaller than 20 ms, which indicates that

the infrastructure of Akamai is effectively distributed.

Observing the average values of RTT shown in Figure

6(c), we can assert that YouTube, thanks to the CDN

created by Google, is globally distributed, bringing

the contents as close as possible to the client. How-

ever it can be noticed a high value of RTT for Asian

countries. Concerning China, the result can probably

be related to the censorship operations implemented

by the Government.

6 DISCUSSION

6.1 Geolocation IPv4 Servers

We apply the techniques described in Section 2 to ge-

ographically locate the servers of video hosting ser-

Internet Streaming and Network Neutrality: Comparing the Performance of Video Hosting Services

519

vices. We have compared the results obtained using

two different techniques: Geoping, based on the val-

ues of RTT measured in every country at each video

download, and Geotrack, which uses traceroute, that

provides the names and IPv4 addresses of the routers

through which the data ﬂow travels from client to

server. Notice that traceroute does not always pro-

vide all the hops for the entire path. The whole op-

eration is affected by issues well known in literature,

for example: load balancing (Augustin et al., 2011),

anonymous routers (Gunes and Sarac, 2008), hidden

routers (Marchetta and Pescape, 2013), misleading in-

termediate delay (Marchetta et al., 2014), and third-

party addresses (Marchetta et al., 2013). For space

constraints we cannot provide all the results obtained.

However, it is possible to summarize the following

results for each providers:

• Dailymotion deploys its entire infrastructure in

France (i.e Paris), no other caches are distributed

elsewhere in the world. However, there are ab-

normal activities by some nodes as we already de-

scribe in the previous section.

• Vimeo has distributed cache-servers, uses the

Akamai infrastructure, and every time a video is

requested, the user is redirected to the “closer”

server. This is highlighted by the lower values

of the RTT and by the names of the servers con-

taining the video (owned by Akamai). The clients

are always re-directed to the same server, without

considering the day time or the network overload,

and only in case the content is not present in the

cache-server, the client is re-directed to the back-

end server.

• YouTube presents its own cache-servers in al-

most all the countries in which there are Plan-

etLab nodes used for testing. The infrastructure

fell within the Internet Exchange Point, in which

they connect to networks via peering to local

ISPs. Unlike Akamai, there is delivery strategy

that assesses both the “distance” between client

and server, and the “overload” of the network.

6.2 IPv4 Identiﬁcation and Name-server

of the Providers

We used the reverse DNS in order to determine the

name-servers associated to the IPv4 of all the servers

contacted. From the list of name-servers conse-

quently obtained, we can say that:

• For Dailymotion, there are only 8 servers from

which clients, from all over the world, download

videos.

• For Vimeo, the servers from which the downloads

are predominantly made, are part of the network

of Akamai

• For YouTube, the servers are globally distributed,

but not always they belong to Google.

Sometimes name-server clearly identiﬁed server

as part of telecommunication or hosting companies,

such as Tiscali, Asianet Web, or Oneandone. Refer-

ring to the discussions on the Network Neutrality and

to the study of IPv4 and name-servers, there are evi-

dences of preferential treatments related to the video

categories. The different performance observed are

due to different infrastructures used by providers. An

interesting case is the Russia where nor CDN or third

party servers are present and all videos coming from

Sweden.

7 CONCLUSION

The aim of our study was to compare the performance

indicators of video hosting services, to understand

whether the performance differences could impact

network neutrality and users privacy. It is worth not-

ing that we do not want to determine whether neutral-

ity is good or not, but we want to evaluate the effect

of the performance differences from the user point of

view, which is of interest for people concerned with

legal and regulatory issues of web applications and

services. We proposed a methodology that, regard-

less the type of provider, allows to acquire and ana-

lyze performance data and qualitative considerations

about the infrastructure of the providers. To vali-

date the methodology, a comparison of the three video

hosting services (Dailymotion, Vimeo and YouTube)

was performed on basis of: performance indicators

(i.e. throughput, RTT and TTL), geography location

of the infrastructures, and routing policies used by the

video hosting services. Results show that Dailymo-

tion seems to have a centralized infrastructure. More-

over, its performance decays with the client’s distance

from infrastructure location. Vimeo and YouTube use

CDNs to deliver their contents, where the ﬁrst showed

the best performance indicators compared to its com-

petitors. Both infrastructures are connected to the

PlanetLab nodes, used as client, often by only two

intermediate hops. We clearly showed that providers

that using distributed infrastructure are actually able

to reach better performance. Regarding network neu-

trality, no evidences of special treatment based on

video category have been collected. The highlighted

performance differences can be regarded as lack of

Server not part of Akamai were also noticed.

ICISSP 2016 - 2nd International Conference on Information Systems Security and Privacy

520

neutrality because all providers should be able to

beneﬁt from the same conditions of distribution and

spread of their contents. However, such differences

are not due to different treatments of trafﬁc, but rather

to different technology infrastructure. Deciding on

whether this is or not a neutrality violation is out of

the scope of this paper. We rather aimed at providing

the regulator with information about the current situ-

ation and performance of video hosting services over

the Internet.

ACKNOWLEDGEMENTS

This work is partially funded by the MIUR project art.

11 DM 593/2000 for NM2 srl.

REFERENCES

Akamai (2012). Improving online video quality and accel-

erating downloads. Accessed: 2015-11-09.

Akamai (2014a). Adaptive media delivery. Accessed:

2015-11-09.

Akamai (2014b). Download delivery. Accessed: 2015-11-

09.

Akamai (2014c). Video streaming services. Accessed:

2015-11-09.

Augustin, B., Friedman, T., and Teixeira, R. (2011). Mea-

suring multipath routing in the internet. Networking,

IEEE/ACM Transactions on, 19(3):830–840.

Calder, M., Fan, X., Hu, Z., Katz-Bassett, E., Heidemann,

J., and Govindan, R. (2013). Mapping the expansion

of google’s serving infrastructure. In Proceedings of

the 2013 Conference on Internet Measurement Con-

ference, IMC ’13, pages 313–326, New York, NY,

USA. ACM.

Cha, M., Kwak, H., Rodriguez, P., Ahn, Y.-y., and Moon,

S. (2007). I tube, you tube, everybody tubes: analyz-

ing the world’s largest user generated content video

system. Proc ACM Internet Measurement Conference

IMC San Diego CA October 2007, New York:1–14.

Chun, B.and Culler, D., Roscoe, T., Bavier, A., Peterson,

L., Wawrzoniak, M., and Bowman, M. (2003). Plan-

etlab: An overlay testbed for broad-coverage services.

SIGCOMM Comput. Commun. Rev., 33(3):3–12.

Cisco (2015). Cisco visual networking index: Forecast and

methodology, 20112016. Accessed: 2015-11-08.

EMC2 (2010). A breakthrough in web-based video content

sharing. Accessed: 2015-11-09.

Google (2015). Google peering & content delivery - content

delivery ecosystem. Accessed: 2015-11-08.

Gunes, M. and Sarac, K. (2008). Resolving anonymous

routers in internet topology measurement studies. In

INFOCOM 2008. The 27th Conference on Computer

Communications. IEEE.

IATA (2010). Iata airport code. Accessed: 2015-11-09.

Lederer, S., M¨uller, C., and Timmerer, C. (2012). Dynamic

adaptive streaming over http dataset. In Proceedings

of the 3rd Multimedia Systems Conference on - MM-

Sys ’12, number ISBN: 978-1-4503-0518-1, page 89,

New York, New York, USA. ACM Press.

Lemley, M. A. and Lessig, L. (2000). The end of end-to-

end: Preserving the architecture of the internet in the

broadband era. UCLA Law Review, 48.

Marchetta, P., Botta, A., Katz-Bassett, E., and Pescap, A.

(2014). Dissecting round trip time on the slow path

with a single packet. In Faloutsos, M. and Kuz-

manovic, A., editors, Passive and Active Measure-

ment, volume 8362 of Lecture Notes in Computer Sci-

ence, pages 88–97. Springer International Publishing.

Marchetta, P., de Donato, W., and Pescap, A. (2013). De-

tecting third-party addresses in traceroute traces with

ip timestamp option. In Roughan, M. and Chang,

R., editors, Passive and Active Measurement, volume

7799 of Lecture Notes in Computer Science, pages

21–30. Springer Berlin Heidelberg.

Marchetta, P. and Pescape, A. (2013). Drago: Detecting,

quantifying and locating hidden routers in traceroute

ip paths. In INFOCOM, 2013 Proceedings IEEE,

pages 3237–3242.

Orange (2014). Orange business services helps dailymo-

tion optimize live streaming for its partners. Accessed:

2015-11-09.

Padmanabhan, V. N. and Subramanian, L. (2001). An

investigation of geographic mapping techniques for

internet hosts. SIGCOMM Comput. Commun. Rev.,

31(4):173–185.

Pantos, R. and May, W. (2015). Http live streaming. Ac-

cessed: 2015-11-09.

Pelaprat, P. (2007). Dailymotion construit son infrastructure

sur un cluster. Accessed: 2015-11-09.

Plissonneau, L. and Biersack, E. (2012). A longitudinal

view of http video streaming performance. In Pro-

ceedings of the 3rd Multimedia Systems Conference,

MMSys ’12, pages 203–214, New York, NY, USA.

ACM.

Plissonneau, L., Biersack, E., and Juluri, P. (2012). Ana-

lyzing the impact of youtube delivery policies on user

experience. In Proceedings of the 24th International

Teletrafﬁc Congress, ITC ’12, pages 28:1–28:8. Inter-

national Teletrafﬁc Congress.

Svensson (2007). Comcast blocks some subscriber internet

trafﬁc, ap testing shows. Accessed: 2015-11-05.

Vimeo (2013). Cdn usage. Accessed: 2015-11-09.

Wu, T. (2003). Network neutrality, broadband discrimina-

tion. Journal of Telecommunications and High Tech-

nology Law, 2:141.

Internet Streaming and Network Neutrality: Comparing the Performance of Video Hosting Services

521