With increasing demand for videos, streaming media, and for services such as cloud computing in Africa, broadband performance, and specifically how users experience performance, becomes increasingly important. In order to meet a growing demand for digital content, mobile operators across the continent have extensively invested in increasing capacity by investing in undersea cables, as well as in terrestrial fibre networks. Mobile LTE networks provision is expanding as well but remains insufficient to cover remote and rural areas.

1. Internet development in Africa: a content use,
hosting and distribution perspective
Enrico Calandro
Research ICT Africa
Josiah Chavula
AfriNIC
Amreesh Phokeer
AfriNIC
With increasing demand for videos, streaming media, and for services such as cloud
computing in Africa, broadband performance, and specifically how users experience
performance, becomes increasingly important. In order to meet a growing demand for digital
content, mobile operators across the continent have extensively invested in increasing
capacity through investing in undersea cables, as well as in terrestrial fibre networks. Mobile
LTE networks provision is expanding as well, but remains insufficient to cover remote and
rural areas.
As more resources and applications are stored and accessed through the cloud, public and
private organisations remotely lease storage and computational resources as needed. For
such remote access to computational resources to work efficiently, there is need for a well
functioning Internet infrastructure to support reliable delivery of remotely hosted content,
services, and applications. Nevertheless, Internet performance on the continent is largely
characterised by slow download speeds and high delays. On the other hand, engineering
problems in Africa related to Internet topology affect performance. Specifically, lack of direct
interconnection amongst Africa’s Internet Service Providers (ISPs), results in suboptimal
performance for intra-country and cross-border communication, as well as high cost of
Internet access. Last but not least, inefficient Domain Name System configurations, as well
as lack of local content caching servers across the African continent are other factors of poor
broadband performance.
By untangling the complexity of content access, use, hosting and distribution in Africa, this
study offers three main contributions. First, it discusses challenges related to usage, hosting,
distribution of local content and services in Africa, by developing a case on African local
news websites. Second, it makes publicly available measurement data and indicators for
local content use, hosting, and distribution across all African countries. And last but not least,
it provides points of policy recommendations on how to improve internet access and use,
and infrastructure performance from a content perspective.
Although a considerable investment in broadband infrastructure has improved broadband
speeds across many African countries, the reliability and performance that users ultimately
receive is determined also by the interconnection between ISPs and by where the content,
services and applications are hosted. Often, high latencies to destinations introduce
significant performance bottlenecks, suggesting that, in addition to investments in higher
throughput links, effort should be devoted to improving interconnection between ISPs and
locating content closer to users.
1

2. Keywords: local content, web hosting, latency, peering, content infrastructure
2

3. Introduction
As a result of new investments in backbone, backhaul and notably wireless access
infrastructure (Song, 2017), Internet availability has outpaced adoption in Africa (Kende and
Quast, 2016), raising questions of why adoption is lagging behind.
The majority of the African population continues to be offline due to high data costs (RIA,
2017a, 2017b, 2017c; A4AI, 2017) lack of local content (Amos, 2016), and poor network
performance (Chetty et al., 2013; Fanou et al., 2014), despite a number of investments and
projects to expand and upgrade u​ndersea cables, and new investments in terrestrial fibre
network capacity. Access to spectrum remains one of the main bottlenecks to network
growth on the continent, and very little progress was made over the last few years to
effectively release 3G and 4G spectrum (Song, 2017).
Between 2013 and 2017, Africa experienced the most rapid growth of international internet
bandwidth in comparison to other regions, growing at a compound annual rate of 44 percent.
(Rebatta, 2017). While this significant investment in broadband infrastructure in Africa has
improved throughputs across the continent, the average Round-Trip Times (RTT or latency)
is still high, due to poor Internet peering infrastructure (​Chavula et al., 2017). Not only is the
peering fabric of the continent uneven, but also ​content infrastructure in Africa requires
significant development. Studies suggest that content is a dominant component of network
traffic, but local content is a major bottleneck to African connectivity (Bruegge et al., 2011).
Research conducted by Fanou et al., (2016), which explored the deployment of web
infrastructure in Africa by surveying 18 African websites, showed that a large number of
regional websites in Africa have their hosting servers outside Africa. ​Even though the
number of data centers across African countries is increasing (Booth, 2014; Jones, 2014),
together with a growing number of Internet eXchange Points (IXPs) to exchange local traffic
1
, along with international content providers and Content Delivery Networks (CDNs) beginning
to install nodes in Africa, most content, even local websites, are hosted and are delivered
from overseas (Bram, 2015). ​As an illustrative example, almost all the Alexa top 50 websites
by African countries are foreign websites . In terms of data centres, Google ​reports to have
2
nine of them in the United States, four in Europe, two in Asia, and none in Africa .
3
A number of factors have been postulated to be the reason for remote hosting of African
websites. Foremost, hosting services and infrastructure have not been pervasive in Africa,
rendering the provision and management of web content within the continent unattractive
1
The updated map on African IXPs is available at the following link: http://www.af-ix.net/ixps-map
2
Although Alexa Top List is one of the mostly used samples for internet measurements studies, the
List might have potential biases, as the user base of the Alexa browser plugins is not representative
(​Scheitle​ et al., 2018).
3
​https://www.google.com/about/datacenters/inside/locations/index.html​. ​Conversely, in order to bring
content closer to the users, Google has de​ployed a number cache infrastructure by deploying
Google-supplied servers inside local network infrastructures and Internet service providers. See also
research by Gupta et al., 2013, on how hosting cache nodes in a local AS reduces latency by more
than 300 ms round-trip times.
3

4. (Kende, 2015). Although it is generally cheaper to host with remote/foreign companies than
with local hosting companies​, content hosted abroad must be routed back to the country of
origin over international Internet transit links that, in spite of significant infrastructure
investments in recent years, are still expensive. The resulting high costs to access content
hosted abroad are generally borne by Internet Service Providers (ISPs), and ultimately by
Internet users. The result is a negative externality, where the economic decisions of content
providers to host abroad have a negative impact on ISPs’ costs, which in turn increases the
cost of Internet usage and limits Internet content demand (Kende & Rose, 2015).
Furthermore, due to topological inefficiencies in many parts of Africa’s Internet, there are
high inter-AS and inter-country delays in many African countries (​Formoso et al, 2017​). As a
4
result, it tends to be more efficient, in terms of RTT, to fetch websites from North America or
Europe than fetching them locally.
From a more social perspective, lack of content in local African languages further reduces
the ability to access and use the Internet. Although social networking platforms, educational
services, and entertainment are relevant in many countries worldwide, content must be in
familiar languages to be relevant, which is often an issue in Sub-Saharan countries whose
populations are not always comfortable in the official government language. While one might
assume this mainly impacts international content, it is also true for local content, including
e-government services, as most of the time they are not offered in local languages.
Early advocates of the internet’s democratizing power believed that the web would give more
people a voice to better participate in their own communities and countries. The web has
long been considered an open and participatory platform (Lessig, 2013; Benkler (2007);
Bruns (2008); Jenkins (2006); Tapscott and Williams (2006); Shirky (2011); in Ballatore,
Graham, and Sen, 2017). Nevertheless, while access has improved, wealthier and better
connected countries create and host the majority of internet content. For instance, research
conducted by Graham, Hale, and Stephens (2011), has shown how Wikipedia’s
user-generated content not only largely represents Global North views but is also
overwhelmingly produced by users in the Global North.
A study by Ballatore et al., (2017) by focusing on search results generated in the 188
countries where Google was available, when searching for capital cities revealed that only
eight African countries have a majority of content that is locally produced. Most of the
5
content comes from the United States (US), and to a lesser extent, from France. Ballatore et
al. (2017) refer to this phenomenon as “digital hegemony”, whereby producers in a few
countries define what is consumed by others.
Financial and skill barriers are only some of the factors affecting who is able to participate in
the digital representations of the world and who does not (Elwood, ​2006​). Other factors, in
the African context, are related to the Internet topology which is characterized by limited
4
An Autonomous System (AS) is a collection of IP routing prefixes under the control of one or more
network operators. It refers to network operators.
5
The African countries that dominate results in local versions of Google are South Africa,
Madagascar, Malawi, Tanzania, Uganda, Code d’Ivoire, Burkina Faso, Togo, Senegal, and Tunisia.
4

5. national and international interconnections and peering. Low levels of Internet penetration
and disposable incomes (​ITU, 2017)​, have been a disincentive for businesses that optimize
the distribution of web content through deployment of Content Delivery Networks (Gupta et.
al., 2014, Fanou et. al., 2015) to invest in Africa.
While these studies on bottlenecks to Internet adoption in Africa have focused on local
content production and usage, and topology issues, little attention has been given to the
local web content hosting fabric and performance issues of accessing such web content.
This study adopts a content use, hosting and distribution perspective to assess the level of
internet development in Africa. It does so by providing empirical evidence on the important
issues of local content by adopting an internet measurements perspective. The research
identifies bottlenecks related to content hosting in African, and therefore provides points of
policy recommendations on how to improve local web hosting across Africa.
Research questions
This paper seeks to ​untangle the complexity of content use, hosting and distribution in
Africa​. ​More specifically, it poses the following key questions:
1. What type of content do African people consume?
2. How much does it cost to access content?
3. Where is local African content hosted? Taking into account local news websites, how
much of the content is hosted locally vs globally?
4. How is content hosted in Africa?
5. What routes are used to access locally hosted content?
6. What is the latency for content hosted in various regions?
Contributions of the study
In answering these key research questions, this paper makes three contributions: First, it
offers a discussion on challenges related to usage, hosting, distribution and accessing of
local content in Africa. Secondly, this study makes publicly available measurement data the
web content infrastructure in Africa, and at the same time it illustrates what the factors
affecting performance when accessing Africa’s digital content are. And last but not least, the
study provides specific points of policy recommendations on how to improve Internet
adoption and infrastructure performance from a content perspective. To achieve this, the
study undertakes an active Internet measurements campaign to characterize the latencies
and to geolocate web servers and routes used for Africa’s online content, focussing on local
news websites in each country. ​This paper explores the hosting patterns and performance
associated with a large sample of about 1,100 local news websites. ​To test assumptions on
content use from a user’s perspective, it draws on nationally representative ICT access and
use surveys conducted by Research ICT Africa (RIA) in 2017 in seven African countries, and
on the RIA mobile pricing (RAMP) indices portal.
5

6. Research Methods and Data Sources
The study makes use of Internet measurements, household surveys data and pricing data as
the three main data sources.
First, in order to measure what type of content people in Africa consume, the study draws
on the Research ICT Africa (RIA) #AfterAccess survey , which delivers nationally
6
representative results for households and individuals. The survey is based on enumerator
areas (EA) of national census sample frames as primary sampling units ​
. Through the
7
survey, we could establish what type of content African people consume.
Second, Internet measurements were conducted to gather information regarding where
Africa’s web content is hosted, as well as to assess the associated performance and cost
implications. The first task was therefore to identify websites that would be considered
representative of Africa’s local web content. In this study, Africa’s web content is defined as
content that is primarily generated and consumed within each African country. It was
decided therefore to study local news and media websites, which by definition constitute a
significant body of local content in Africa. A list of local news websites made for every
African country was compiled from ​ABYZ News Links , an online directory of links to online
8
news sources from around the world organized on a geographical basis .
9
To assess how much it costs to access African news websites, two web-crawlers, one
mobile-device based and another desktop based, were used to scrape websites’
homepages. Each website’s index page is examined by downloading all the web objects on
the home page. This makes it possible to compute the amount of data downloaded by
consumers each time they visit the websites. The price of mobile Internet access (per
gigabyte of data) , as well the sizes of the websites’ homepages (index page), provide an
10
indicator on how much it costs to access websites.
6
https://researchictafrica.net/2017/08/04/beyond-access-surveys-questionnaires-methodology-and-tim
eframe/
7
The random sampling is performed in four steps for households and five steps for individuals. First,
the national census sample frames is split into urban and rural enumerator areas (EAs); second, EAs
are sampled for each stratum using probability proportional to size (PPS); third, for each EA a listing
of households is compiled. The listing serves as sample frame for the simple random sections of
households; forth, a specific number of households is sampled using simple random sample for each
selected EA; and fifth, from all household members 15 years or older or visitors staying the night at
the house, one individual is randomly selected based on simple random sampling.
8
Available at the following link: http://www.abyznewslinks.com/
9
One way of sampling websites is through listings of the most popular websites as ranked by
webometric sites, the most popular one being the Alexa Top. Our analysis of the Alexa top-50 sites for
African countries showed that lists were largely dominated by non African sites. It was also noted that
some local websites that are anecdotally known to be very popular in some African countries were
missing as top websites, thereby casting some doubts on the representativeness of the webometrics
for Africa. In contrast, the ​ABYZ News Links ​directory does not rank the sites, but rather attempts to
list all the prominent media sites for each country.
10
1GB basket data are extracted from the RIA African Mobile Price Index (RAMP). RAMP is a comprehensive
mobile pricing database of Africa. The data is collected quarterly and covers all African operators from all African
countries.
6

7. Third, to answer to the question of whether local news websites are locally hosted within
their countries or not, Traceroute data was analysed to determine the networks that host
each of the measured websites, as well as the networks through which traffic flows between
the websites and the measurement vantage points. Subsequently, the geographical location
of each web-hosting server was determined. MaxMind geolocation database was used to
11
obtain the network information, which includes the networks’ Autonomous System Numbers
(ASNs) and network names. The geolocation database was also used to identify the
countries related to each IP in the dataset, both the websites’ web-servers and routers
along the paths to the websites. The country-level geolocation was prefered as it has been
shown to have relatively higher accuracy compared to city-level geolocation (​Poese et. al.,
2011, Shavitt et. al., 2011) .
12
Fourth, the key inquiry on each country’s news websites was to determine the countries and
networks in which they are hosted, as well as the attendant packet delays between users in
the country and websites. To obtain this information, active measurements, in the form of
Traceroute, were conducted from a distributed set of vantage points in each country to the
respective websites. A common platform for performing distributed active measurements is
the Ripe Atlas , which provides a world-wide network of physical probes that can be used
13
as measurement vantage points. Ripe Atlas presently has about 230 active probes in Africa,
distributed across 36 countries.
Data Characterisation and Overview
Survey data
Data on internet use across African countries were collected from in-depth individual surveys
conducted in 2017 in the following seven countries: Ghana, Kenya, Mozambique, Nigeria,
Rwanda, South Africa and Tanzania. The survey data includes individual and household
level information on mobile phone and Internet access and use. As a result of data sampling
process, a total of 9 163 respondents participated in the survey. Table 1 below shows the
number of individual surveyed in each country and the share of male and urban areas in the
whole sample. Once the data is weighted to correct for over- or under representativeness of
rural/urban, age groups, there is evidence that the majority of the population reside in rural
areas except for Ghana and South Africa.
TABLE 1: SAMPLE DISTRIBUTION
Country Observation Urban (%)
Ghana 1200 55,31
11
https://www.maxmind.com/en/geoip2-country-database
12
It is worthy pointing out the known limitations with geolocation databases; it is likely that
discrepancies within MaxMind introduce noise to the data analysis. However, as stated above, the
analysis in this study is limited to country-level geolocation to minimise impact of database
inaccuracies.
13
https://atlas.ripe.net/
7

8. Kenya 1208 26,49
Mozambique 1171 32,82
Nigeria 1200 49,4
Rwanda 1211 21,6
South Africa 1815 64,5
Tanzania 1200 33,02
Source: Research ICT Africa #afteraccess survey, 2017
Internet measurement data
The ​ABYZ directory provided a list of about 1,065 African news sites, distributed among the
different countries as summarized in Table 2. From each country and for each website, a
maximum of 10 probes were selected and used to launch Traceroute packets to each of the
country’s websites.
Table 2: Number of news websites measured per country
AO BI BJ BW CD CG CI CM CV DZ EG ET GH GM KE LS MA MG
26 12 23 13 22 8 22 28 9 100 26 27 44 10 23 5 37 10
MU MW MZ NA NG RE RW SC SD SN SS SZ TG TN TZ UG ZA ZM ZW
10 19 7 15 176 5 8 4 11 21 13 3 18 19 18 31 126 38 78
The Traceroute measurements were repeated over a five day period, resulting in about
19,299 successfully measurements between the probes and the websites. Each Traceroute
measurement returns three final hop RTTs, meaning that in total, there were 57,897
end-to-end RTTs. A Traceroute measurement is considered successful if an IP route can be
determined from the source to the web-hosting network, thereby also being able to reveal a
delay estimate to the website. Each successful measurement contains the IP address of a
website’s hosting server, a series of IP hops from the vantage point up the server, as well as
the delays (round-trip time, RTT) at each hop (router). Also, each Traceroute result is made
up of multiple records, one record for each of the multiple hops on the path. Consequently,
the final dataset was made up of 256,654 records, with each record comprising source and
destination addresses, as well as the IP hop and RTT from the source to that hop.
Table 3: RTT samples
Country Measurements
Completed
RTT Samples
8

9. BI 102 306
BJ 580 1740
BW 160 480
CD 211 633
CG 40 120
CI 172 516
CM 823 2469
CV 45 135
DZ 947 2841
EG 181 543
ET 257 771
GH 547 1641
GM 98 294
KE 767 2301
LS 25 75
MA 619 1857
MG 177 531
MU 341 1023
MW 368 1104
MZ 234 702
NA 148 444
NG 1381 4143
RE 236 708
RW 38 114
SC 32 96
SD 107 321
SN 308 924
SS 64 192
SZ 14 42
TG 651 1953
TN 801 2403
TZ 662 1986
UG 564 1692
ZA 5284 15852
ZM 465 1395
ZW 1495 4485
Total 19299 57897
Data Analysis
9

10. Internet users perspective
Table 4 below shows that mobile phone penetration in African countries has not reached the
100% as reported by ITU for some African countries (ITU, 2017). The mobile phone
technology continues nevertheless to scale rapidly with more than 50% of the African
population owning a mobile phone. Migration to higher speed networks and smartphones
continues apace, with mobile broadband connections set to reduce the historical digital
divide. In four of the surveyed countries, more than 20% of respondents have used the
Internet, but in Mozambique (9,70%), Rwanda (8,21%) and Tanzania (13,53%), which
constitute the poorest of the countries surveyed it is below 20%.
As the vast majority of people in all seven countries access the Internet through their mobile
phone, the low Internet penetration in these countries can be attributed to low smartphone
penetration which, except for Tanzania, is lower than 20% compared to South Africa
(55,53%), Ghana (34,27%), Kenya (27,57%) and Nigeria (23,83%). Surprisingly, Tanzania’s
smartphone penetration is above 20% but Internet penetration remains lower. This could be
attributed to supply-side issues such as data prices or the dearth of skills to enable Internet
use.
Table 4: Mobile phone, smartphone, and internet use across 7 African countries
Country Mobile Phone (%) Smartphone (%) Internet Use (%)
Ghana 73,87 34,27 26,00
Kenya 86,94 27,57 25,59
Mozambique 39,73 17,01 9,70
Nigeria 64,42 23,83 30,22
Rwanda 48,16 9,02 8,21
South Africa 83,84 55,53 49,72
Tanzania 58,52 22,12 13,53
Source: #AfterAccess RIA surveys, 2017
Across four of the countries surveyed , lack of content does not seem to be one of main
14
reasons preventing people from using the mobile phone. Issues of cost have been identified
as the main barriers to mobile phone usage. In Rwanda (11.55%), users expressed a
concern related to sending their personal data through the phone, while in South Africa, after
cost of calls (34.33%) and cost of data (20.11%), coverage (6.51%, most probably in rural
areas) has been reported as a major problem.
14
Unfortunately, some of the questions of the #AfterAccess surveys were not asked across all the
countries under investigation.
10

11. Table 5: ​Reasons preventing from using the mobile phone more
Mozambique Rwanda South
Africa
Tanzania
Friends and family do not have mobile
phone
8,69 5,04 2,56 0,64
Cost of calls (airtime) 54,64 43,39 34,33 51,92
Cost of data 3,7 5,03 20,11 5,1
Coverage (available services) 3,99 1,41 6,51 6,71
Battery life 11,28 2,15 4,54 16,45
Lack of content 0,85 6,12 1,66 1,75
My mobile phone is a distraction 1,17 3,28 2,48 8,51
I am worried about sending personal
information
1,28 11,55 1,6 1,01
Not applicable (I do not want to us 3,19 15,63 21,96 1,96
Other 11,2 6,4 4,25 5,95
Source: #AfterAccess RIA surveys, 2017
Regarding Internet use, African users spend most of their time on social media (58.49%),
followed by education activities online (20.38%). For those accessing to the internet,
work-related activities are also predominant. It is worth to note that although news and
entertainment were not mentioned as some of the most time-consuming activities online,
most of the social-media content is actually about entertainment and news.
Table 6: ​When you use the internet, what do you spend more time on?
Kenya Mozambique Ghana Nigeria South Africa Tanzania
Work 16,13 10,97 7,27 16,7 19,53 9,17
Education 9,87 17,4 31,72 20,63 24,94 17,7
Social media (like
Facebook)
66,38 67,71 52,64 52,43 44,07 67,7
News 3,24 2,56 1,34 5,91 2,74 2,53
Entertainment 0,98 0,93 4,62 1,36 3,91 1,92
Other 3,4 0,44 2,41 2,97 4,82 0,98
11

12. Source: #AfterAccess RIA surveys, 2017
In terms of barriers to internet use (Table 7 below), as expected, data cost has been
reported as the main obstacle. For a number of respondents, internet is considered a
time-consuming activity, as it seems that lack of time is another relevant barrier to internet
use. In Tanzania (28.36%), South Africa (24.22%), and Mozambique (36.5%), instead, the
users perceive that internet speed is not sufficient for a seamless internet access. Lack of
content in local language, on the other hand, and in contrast to previous studies on local
content in Africa, is not considered one of the main obstacles to internet use, except in
Rwanda, where 8,49% of the respondents expressed some concern related to lack of
content in local languages.
Table 7: ​What does limit you from using the internet?
Ghana Kenya Mozambi
que
Nigeria Rwanda South
Africa
Tanzani
a
No limitation 11,96 16,95 19,92 21,58
Lack of time 21,78 20,16 11,59 15,65 18,13 10 25,62
Data cost 51,51 45,42 43,28 32,25 48,7 47,15 40,64
Lack of content in
mhy language
3,59 1,96 6,43 0,26 8,49 3,32 3,68
Speed of Internet 7,53 11,63 36,5 18,11 1,01 24,22 28,36
Privacy concern 0,47 0,51 2,44 2,98 2,08 3,18 0,89
Worried about
getting
virus/malware
0,6 0,74 8,85 9,97 3,77 0,85
Not allowed to use
it (by
family/spouse)
0,69 0,07 5,45 1,02 2,95 2,88 0,57
Find it difficult to
use
1,87 0,95 5,36 1,65 5,99 2,23 4,14
Cost of accessing websites
As depicted in Figure 1 below, about half of the websites have homepage data volumes
exceeding 1 MB. It is interesting to note that the size of websites’ homepages determined
using mobile phone browser user-agent, are almost similar in size to homepages obtained
using a desktop browser. This does suggest that most of the news websites are not
optimized for mobile access.
12

13. Figure 1: Size of news website home pages, calculated based on sum of the size for web
object downloaded when loading site homepages.
As Figure 2 below shows, the size of home pages has a considerable bearing on the
comparative cost of accessing a news website. For more than one third of all African
countries under investigation, accessing the homepage costs more than USD0.01. The most
expensive countries are Congo, Ghana Bissau, Zimbabwe, Lesotho, Niger, and Gambia.
Figure 2: Cost of accessing news website home pages, calculated based on size of
homepage (MB) multiplied by the average data cost in a country.
Geolocation of African news content hosting
The hosting and geolocation analysis indicates that about 85% of the news websites are
hosted outside the countries in which they belong, i.e. the website is owned and it is local to
one country, but is hosted in another country. This is, hereafter, referred to as remote
hosting. ​Analysis of remote hosted websites reveals that most of them are hosted in Europe
and the US. The chart in Figure 3 shows the distribution of countries acting as remote hosts
for African websites. ​This signals low participation of the continent’s companies in content
hosting. ​Most of the websites that were observed to be hosted within Africa were based in
South Africa, while the majority of the rest are hosted in either the US or Europe.
13

14. Figure 3 below shows a country-level distribution of locally hosted websites versus remotely
hosted. Almost all the countries in the sample have less than 30% of their websites hosted
locally, and about half of all the countries have less than 10% local hosting. South Africa
(ZA) and Swaziland (SZ) appear to have the highest percentage of local hosting; 46% for
South Africa and 65% for Swaziland.
Figure 3:​ Percentage of websites per country local vs remote
Figure-4: ​Map of the world showing countries where the African news websites are hosted;
the color intensity reflects the percentage of websites hosted in the country.
14

15. From Figure 5 below it is possible to observe that the US takes the lion’s share in hosting
African content, with about 58% of all the websites being hosted by American companies.
Within Africa, South Africa leads in the content hosting business, hosting about 14% of all of
Africa’s remotely hosted news websites (i.e minus those that belong to South Africa). The
rest of the websites, about 20% are hosted in various countries in Europe (notably, 9% in
France, 4% in Germany, and 3% in Great Britain).
Figure 6 below shows that about 45% of all the IP hops (i.e. Internet path) for accessing
African websites from African countries traversed outside African clients’ home countries.
Internet packets travel mostly through US networks, and about 23% pass through European
networks. South Africa takes about 8% of all IP hops for traffic traversing to other African
countries.
Figure 5: Country-level distribution of
remotely hosted African news websites
Figure 6: Country-level distribution of
Internet paths (IP hops) for accessing
African websites from African countries.
Network-level analysis of Africa’s news sites
Similar to the geo-location analysis, network-level analysis shows that most of the websites
are hosted by foreign companies. Chart in Figure 7 below shows the distribution of websites
among the networks. Taking into consideration all the sampled African news websites,
Cloudflare Inc (US) takes the biggest share of the market, hosting about 22% of the
websites. Following in the far distance is OVH SAS (France) with 8%, OPTINET (South
Africa) at 6%, Google LLC and GoDaddy.com (both US) at 5% each, and Unified Layer (US)
at 4%, and HETZNER (South Africa) at 3%. Figure 8 depicts the hosting market share if only
remotely hosted websites are considered, Cloudflare take an even bigger share of 26%,
followed by OVH SAS (9%), Google LLC (6%), GoDaddy.com (5%) and Unified Layer (5%).
What is interesting to note is that the leading providers for Africa’s remote hosted news
websites are largely based on Cloud infrastructure and make use of content distribution
networks.
15

16. Figure 7: Hosting distribution of all the
sampled African news websites
Figure 8: Hosting distribution for remotely
hosted African news websites
Delay Analysis (Round Trip Times) to access locally and remotely hosted content
Analysis of round-trip times (RTTs) when accessing the websites from each of the countries
shows significant RTT differences between locally hosted websites and those remotely
hosted. The maps in Figure 9 and Figure 10 below highlights country-level RTT differences
for local and remote hosting. The median RTTs for locally hosted websites is about 50ms
(Figure 9), whereas for remote hosted websites, the median RTTs range between 100ms
and 300ms (Figure 9). This shows that there are significant performance implications for
geo-location of website hosting.
Figure 9: National median RTTs to locally
hosted websites
Figure 10: National median RTTs to
remotely hosted websites
16

17. Figure 11: Median RTTs (ms) between measurement vantage points and news websites
located within each country
Figure 12: Median RTTs (ms) between measurement vantage points and news websites
located in remote countries
The range of median RTTs shows significant differences between countries for websites
hosted within each country. As presented in boxplots Figure 11 above, out of the 22
countries where local hosting performance was measured, 16 countries registered local
median RTTs of less than 50ms, while the other 6 countries (Angola, Malawi, Lesotho,
Algeria, Cameroon, Morocco, and Ghana) had local median RTTs ranging between 50ms
and 200ms. The analysis shows that in some countries, the median RTTs for locally hosted
websites is higher than for websites that are remotely hosted. Examples include Ghana,
where the median RTT for locally hosted websites was found to be 205ms, whereas for
remote websites, the median RTT was 127ms (Figure 11); and Morocco, where the local
median RTT was 152ms against a remote median RTT of 68ms. RTTs of over 100ms for
locally hosted websites could suggest circuitous path, where locally hosted websites are
accessed through Internet paths that traverse other countries. This further indicates a lack
peering, where interconnections between local network is done through networks in remote
17

18. countries. In Ghana for example, one probe reached the locally hosted ​www.ghana.gov.gh
by traversing remote networks, chronologically through Ghana, South Africa, UK, South
African, and Ghana, resulting in a delay of 380 ms. Similarly in Morocco, a locally hosted
website ​www.leseco.ma was reached by a probe in Morocco by traversing four networks,
first in Morocco, then France, Ireland, Canada, and back to Morocco, experiencing a delay of
160ms. These examples illustrate that in the absence of local peering, local hosting of
websites tends to force much more circuitous routes for accessing the content than when the
websites are in foreign countries, resulting higher delays and poor performance for
consumers.
From a performance perspective, it is interesting to make some comparison between
remote websites that are hosted by CDN-enabled networks. In a nutshell, CDN refers to
groups of servers that are distributed in various geographic locations and work together to
provide fast delivery of Internet content. The CDN takes content that is otherwise hosted on
a single server, and replicates it to a set of distributed servers that are deemed to be closer
to the intended consumers of such content. On the other hand, content that is not supported
by CDN infrastructure generally remains within its original server location. This means that,
while locally hosted content will be closer to the local consumers, CDN-based content can
be ​brought closer to consumers even if the original servers are in distant locations. The
expectation, therefore, is that level of delays for CDN-based websites should be similar to
locally hosted websites. Of course, this assumes that the CDNs have nodes within or close
to the respective countries. In this regard, the South African case is worth of a mention.
Although the country has only about 46% of the news websites hosted locally, it can be seen
the median RTTs for local and remote websites are almost the same; 22ms and 25ms
respectively. As was mentioned earlier, the leading remote hosting providers for Africa
(Cloudflare, ​OVH SAS, Google LLC, GoDaddy.com and Unified Layer) ​operate on ​Cloud
infrastructure and distribute their content via CDN services. It is also worth noting that South
Africa hosts a number of CDN nodes, including Cloudflare, which has two; one in Cape
Town and another node in Johannesburg. This means that although the original
15
webhosting is in remote countries, the actual website content is generally served from within
the country. CDNs are helpful primarily to bring content hosted overseas closer to its users,
and the increasing in local traffic might become an incentive for local ISPs to peer locally.
Discussion
The landing of higher speed networks and the increase in smartphones adoption across the
continent is expected to reduce the historical digital divide. However, in 2017, in three out of
seven of the countries surveyed, less than 20% of the respondents have used the Internet.
Considering that the vast majority of people in Africa access the Internet through their mobile
phone, the low Internet penetration can be attributed to low smartphone penetration which,
except for Tanzania, is lower than 20%. While the high cost of airtime has been identified as
the main barrier to mobile phone usage, according to demand-side data lack of content does
not seem to be one of main reasons preventing people from using the mobile phone.
Similarly, in terms of barriers to internet use high costs of data is hindering adoption.
15
https://www.cdnplanet.com/geo/south-africa-cdn/
18

19. Contrary to previous studies on Internet adoption, this study does not find evidence of lack of
content in local languages as one of the main obstacles to internet use, except in Rwanda.
In more than one third of all African countries sampled, accessing local news websites costs
more than USD0.01. Half of these websites have homepage data volumes exceeding 1 MB
and they are are not optimized for mobile access. 85% of the local news websites are hosted
outside of their respective countries, mostly Europe and US, while the majority of the
websites that were observed to be hosted within Africa were based in South Africa. Almost
all the countries in the sample have less than 30% of their websites hosted locally, and
about half of all the countries have less than 10% local hosting. 68% of all the Internet path
for accessing African websites from African countries traversed outside Africa, mostly
through US and European networks. A ​network-level analysis showed that most of the local
African news websites are hosted by foreign companies. The leading providers for Africa’s
remote hosted news websites are largely based on Cloud infrastructure and make use of
content distribution networks to mirror the content locally. ​One direct consequence of remote
hosting is that African network operators have to use significant levels of international
bandwidth to fetch African content for their clients. The cost of this international bandwidth is
in the end passed on to the Internet users in Africa.
Geo-location of website hosting has significant implication on performance. The median
RTTs for locally hosted websites is generally lower than the RTTs for remote hosted
websites. However, in some countries, the median RTTs for locally hosted websites is higher
than for websites that are remotely hosted. High RTTs levels for locally hosted websites is
an indicator of circuitous paths, which are due to lack of local ISP peering. Rather, in these
cases interconnections between local networks is done through networks in remote
countries. In the absence of local peering, local hosting of websites tends to resolve into
circuitous routes for accessing the content than when the websites are in foreign countries,
resulting in higher delays and therefore in poor performance for Internet users. By bringing
remotely hosted content closer to the users, CDN-enabled networks reduce delays for
CDN-based websites. Although level of delays for CDN-based websites are similar to locally
hosted websites, CDNs do not improve the performance of locally hosted content in cases
where there is lack of local peering.
While access network infrastructure has improved in Africa, the content infrastructure has
lagged behind. There is a considerable difference between access infrastructure and content
infrastructure. While supply-side policy and regulatory interventions have positively affected
access infrastructure (notably the roll out of mobile networks in the case of Africa), content
infrastructure has not always been enabled by national policy and regulatory interventions
resulting in most of the content in Africa being hosted and accessed from overseas. This
configuration is not ideal, as it increases latency levels, and costs to access content. Not
having a reliable hosting infrastructure affects also in-country delivery infrastructure. ​Without
the necessary local peering, locally hosted content can exhibit equally higher access
latencies.
Conclusion and Policy Implications
19

20. This study has provided empirical evidence on the current configuration of web content
hosting, access, and distribution in Africa, and has demonstrated that the status of the
African content infrastructure is alarming. All African countries heavily rely on foreign
services, both to host, to access, and to distribute local content in Africa. Latency levels to
remotely hosted local content are high as well as costs of accessing remotely hosted local
content.
Most of the public policy strategies on improving local content in Africa focused on
demand-side interventions, such as the creation of content in local languages, and on
developing skills on web content production and consumption. While these policies are
important, bodies in charge of the governance of the internet are urged to identify ways of
facilitating local markets for content hosting, access and distributions by focusing on:
1. Incentivising investments on data centres and web farms in Africa, to stimulate economies
of scale for the local web hosting market;
2. Encouraging local news websites to move the content closer to the users in Africa, by
incentivising the use of CDN-enabled networks and by reducing prices for local hosting;
3. Facilitating peering relationships between ISPs and investing in local exchange points to
reduce latency; and
4. Incentivising ISPs to peer in local exchange points.
Future research
While this paper has shown that many African countries still rely on remote hosting, and that
there are negative performance implication, it is also apparent that most of the remote
hosting companies serve the content through CDNs. However, not all African countries have
CDNs within or near them. Another angle worth investigating is the geographical locations of
CDN nodes in Africa, and to understand the attendant web performance differences between
the different regions of the continent. Also, considering the inaccuracies of geolocation
databases with respect to Africa, it might also be interesting to consider other means of
geolocation of CDN nodes. One way would be to apply Internet measurements and RTTs to
improve geolocation databases.
Another relevant area of investigation, not covered by this paper, is on the motivations for
internet service providers to host overseas rather than locally. The paper has discussed that
sub-optimal peering agreements might make it expensive for local ISPs to host locally. It is
worth to explore more in detail what the main reasons for that are. This can be achieved
through an online survey targeted for ISPs complemented by in-depth interviews with
selected ISPs.
References
Alliance for Affordable Internet (2017). Africa Regional Snapshot. 2017 Affordability Report.
Available at
http://1e8q3q16vyc81g8l3h3md6q5f5e.wpengine.netdna-cdn.com/wp-content/uploads/2017/
03/A4AI-2017-Africa-Affordability-Report_Online.pdf
20

21. Amos, I. (2016). Is lack of local content hindering adoption in Africa? In: iAfrikan. 21
September 2016. Available at
https://www.iafrikan.com/2016/09/21/is-the-lack-of-local-content-hindering-internet-adoption-i
n-afrika-2/
Ballatore, A., Graham, M., and Sen, S. 2017. Digital Hegemonies: The Localness of Search
Engine Results. ​Annals of the American Association of Geographers​.
DOI:10.1080/24694452.2017.1308240.
Booth, N. (2014). New projects put emphasis on African data centres growth. In:
Datacentredynamics. Available at
http://www.datacenterdynamics.com/content-tracks/colo-cloud/new-projects-put-emphasis-o
n-african-data-center-growth/86995.fullarticle
Bram, U. (2015). Why your internet connection is slow wherever you are in Africa. in: Quartz
Africa. Available at
https://qz.com/472028/why-your-internet-connection-is-slow-wherever-you-are-in-africa/
Bruegge, C., Ido, K., Reynolds, T., Serra-Vallejo, C., Stryszowski, P., Van Der Berg, R., and
Recuero Virto, L. (2011). The relationship between local content, Internet development and
access prices. Internet Society, OECD, UNESCO. Available at
https://www.oecd.org/internet/ieconomy/50305352.pdf
Bruns, A. (2008). ​Blogs, Wikipedia, Second Life, and beyond: From production to
produsage​. New York: Peter Lang.
Chavula, J., Phokeer, A., Formoso, A., and Feamster, N. (2017). An insight into Africa’s
country-level latencies. In ​Proceedings IEEE AFRICON 2017​, Cape Town.
Chetty, M., Sundaresan, S., Muckaden, S., Feamster, N., and Calandro, E. (2013).
Measuring Broadband Performance in South Africa. Available at
http://www1.icsi.berkeley.edu/~srikanth/docs/broadband-sa-dev4.pdf
Elwood, S. (2006). Critical issues in participatory GIS: Deconstructions, reconstructions, and
new research directions. ​Transactions in GIS​ 10 (5): 693–708.
Fanou, R., Tyson, G., Francois, P., & Sathiaseelan, A. (2016, April). Pushing the frontier:
Exploring the african web ecosystem. In ​Proceedings of the 25th International Conference
on World Wide Web​ (pp. 435-445). International World Wide Web Conferences Steering
Committee.
Fanou, R., Francois, P., & Aben, E. (2015, March). On the diversity of interdomain routing in africa. In
International Conference on Passive and Active Network Measurement​ (pp. 41-54). Springer, Cham
21

22. Formoso, A., Chavula, J., Phokeer, A., Sathiaseelan, A., & Tyson, G., ​Dissecting the African
Internet: An Intra-Continental Study, ​viewed 13 October 2017 from
http://www.eecs.qmul.ac.uk/~tysong/files/africa-internet.pdf
Graham, M., S. A. Hale, and M. Stephens (2011).​ Geographies of the world's knowledge​.
London: Convoco!
Gupta, A., Calder, M., Feamster, N., Chetty, M., Calandro, E., Katz-Bassett, E. (2014)
Peering at the Internet's Frontier: A First Look at ISP Interconnectivity in Africa. PAM,
204-213
International Telecommunications Union (ITU) (2017). Measuring the Information Society
Report 2017. Available at
https://www.itu.int/en/ITU-D/Statistics/Documents/publications/misr2017/MISR2017_Volume
1.pdf
Jenkins, H.(2006).​ Convergence culture: Where old and new media collide​. New York: New
York University Press.
Jones, P. (2014). East Africa data center reports rapid market growth. In:
Datacenterdynamics. Available at
http://www.datacenterdynamics.com/content-tracks/colo-cloud/east-africa-data-center-report
s-rapid-market-growth/86962.fullarticle.
Kende, M., and Rose, K. (2015). Promoting Local Content Hosting to Develop the Internet
Ecosystem. January 2015. Internet Society.
Lessig, L. (2003). An information society: Free or feudal. ​Lecture given at the World Summit
on the Information Society​, Geneva,
Switzerland.​http://www.itu.int/wsis/docs/pc2/visionaries/lessig.pdf
Poese, I., Uhlig, S., Kaafar, M. A., Donnet, B., & Gueye, B. (2011). IP geolocation databases:
Unreliable?. ​ACM SIGCOMM Computer Communication Review​, ​41​(2), 53-56.
Research ICT Africa (2017a). Botswana telecommunications limp a decade after policy
changes. Policy Brief 1, February 2017: Botswana. Available at
https://researchictafrica.net/polbrf/Research_ICT_Africa_Policy_Briefs/2017%20Policy%20B
rief%201_Botswana%20.pdf
Research ICT Africa (2017b). South African data prices static for two years but consumers
not flocking to cheapest product offering. Policy Brief no. 3 2017: South Africa.
https://www.researchictafrica.net/polbrf/Research_ICT_Africa_Policy_Briefs/2017_Policy%2
0Brief%203_South%20Africa.pdf
Research ICT Africa (2017c). SADC not bridging digital divide. Policy Brief 6 September,
2017: SADC. Available at
22

23. https://www.researchictafrica.net/polbrf/Research_ICT_Africa_Policy_Briefs/2017_Policy_Bri
ef_6_SADC.pdf
Scheitle, Q., Jelten, J., Hohlfeld, O., Ciprian, L., and Carle, G. (2018). Structure and Stability
of Internet Top Lists. Available at https://arxiv.org/pdf/1802.02651.pdf
Shavitt, Y., & Zilberman, N. (2011). A geolocation databases study. ​IEEE Journal on Selected Areas
in Communications​, ​29​(10), 2044-2056.
Stucke, W., Esterhuysen, A., Masilela, L., Elkins, M., Jensen, M., and Salih, S. (2016). The
2016 African Domain Name System Market Study. Final Report, 6th June 2017. ICANN.
Available at
https://dnsafrica.study/AfricaDomainNameSystemMarketStudy2016FinalReportVersion1.03h
.pdf
Song, S. (2017). The Internet Is U-Shaped. In Many Possibilities. Available at
https://manypossibilities.net/2017/03/the-internet-is-u-shaped/
23