This document discusses considerations for ICT infrastructure in low-resource settings. It notes that the best infrastructure is not always the highest specification technology, and directly copying approaches from high-resource contexts usually does not work. Both centralized and distributed deployment models are discussed, as are examples of technologies that were successful or not. Key factors that are emphasized include using local expertise and materials, considering total cost of ownership, and designing solutions that are sustainable given the local context and needs of users.
2. What to expect from this lecture
● Overview of low-resource infrastructures
● Introduction of some technological solutions
● Highlight of "do"s and "don't"s
5. ICT here and there
● Same data needs and expectations but...
○ Here: push technology, usage will follow
○ There: spot usage and find matching technology
● Direct implications to keep in mind:
1. The best infrastructure is not necessarily
the one with the highest specifications !
2. Cloning what we enjoy here will most
probably not work there !
6. Technology enhanced communication
● Several elements are implied
● Need to pay attention to all of them in their
context
Software Software
Computing device Computing device
Network
8. Two types of deployment
● Bring the ICT to the users, or the inverse
9. Bring the users to the ICT
● Examples
○ Community computer center (Video)
○ Internet café
○ "Computer room" in schools
● Advantages
○ One installation impact many users
○ Centralisation of infrastructural needs
● Disadvantages
○ Concentration of valuable items
○ Typically high installation costs
○ Harder to embed into culture / habits
10. Bring the ICT to the users
● Examples
○ "One Laptop per Child" initiative
○ Smart phones and tablets
● Advantages
○ "Personal" relationship with the device
○ De-centralisation of resource usage
● Disadvantages
○ One device per person
○ Higher deployment costs
11. What's best to do?
● Very much depending on the context
○ Easy access to a central location?
○ What kind of usage for the devices?
● It is possible to hybridise the approaches
○ e.g.: use laptops and print in a cyber-café
● Also think of embedding other
communication channels
○ prints
○ USB sticks
○ ...
12. Use local expertise and materials
● In any case, remember that
○ Devices will have to be serviced on spot
○ Training individuals to operate something is costly
○ Not everything is available everywhere
○ Uptake will be better from involved communities
● Thus, to increase chances of success
○ Source material locally as much as possible
○ Source expertise locally as much as possible
○ Ensure appropriation of what is deployed
● "Colonialism 2.0"
13. Example: baby incubator [1,2]
● Ship baby incubator to
developing countries
● Big failure
○ $40000 per unit
○ Hard to find spare parts
○ Closed design
○ Need skilled staff
14. Example: baby incubator [1,2]
● Build baby incubator
in, and for, developing
countries
● NeoNurture
○ $1000 per unit
○ Car parts
○ Open design
○ Need staff able to repair
and use a car
22. Electricity consumption
● Some of the things to power up
○ Power supply
○ CPUs and GPUs
○ Screens / Lights
○ Printing devices
○ Storage devices
○ Networking devices
○ Cooling devices
● Get electricity from
○ The grid (Solar, Wind, Nuclear, Tidal, ...)
○ Local production (Solar, Fuel, ...)
23. Consumption of a CPU
● Varies a lot depending on the CPU class and
internal architecture (Intel / ARM)
● Indicated through the TDP value
Server Desktop Netbook Smart phone
69 to 150 W 19 to 73 W 0.65 to 13 W 0.5 to 2 W
24. Consumption of a Screen
● Varies depending on the technology
0.4 to 1.7 W
http://pixelqi.com
25. Consumption of storage device
● Essentially depends on the technology
● Not having movable parts is better
26. Using electricity from the grid
● Pluses
○ Stable, and affordable,
source
● Minuses
○ High throughput of
generation point
○ Significant
infrastructure costs
○ Electricity does not
travel well -> hard to
connect remote areas
27. Local production
● Pluses
○ Easy to deploy
● Minuses
○ Less stable than grid
○ High initial costs
28. More about solar power
● Estimated
1000W per m2
● Actual up to
200W per m2
● Better bet on
less in practice
http://www.alternative-energy-news.info/technology/solar-power/
30. Sending a message from A to B
● In theory, like the post : wrap the message
and send it to the address of B
● In practice this is not so simple
○ First,find the address of B
○ Then, send the message through a relay of
machines and connections
● Let's see what happen when "univ-tours.fr" is
typped in a Web browser ...
31. Find the address of "univ-tours.fr"
● This has to be translated into
"193.52.209.12", the actual address of the
machine
● Use the DNS system to do that
○ Ask the root server ".fr" for the address of "univ-
tours"
○ Ask "univ-tours" for its default host address
● Such names are called TLD and have to be
paid for (roughly 15 euros per year)
32. Ask for the Web index page
● The Web browser sends a request for
"index.html" to the port 80
● Goal: send the request from a machine in
the Netherlands to another (probably) in
France
● See what actually happens with the tools
"tcptraceroute" and "traceroute"
33. What happens in practice...
● The Web browser sends a request for
"index.html" to the port 80
●
34. Why ?
● Internet is a network of networks
● Several routes can connect two devices
● Traffic is agglomerated on several levels
● ISPs exist for each level
37. Why does it matter in ICT4D ?
● Let two farmers in Mali send a mail to each
other implies a couple of regional servers
● If they want to use GMail, a backbone
connection to the US' west coast will also be
involved !
● Consequences
○ Web-based social networking can hardly be local
○ The costs for the end user is likely to be high
○ ( Cloud computing is incompatible with ICT4D )
38. Good news! transit prices are falling
http://www.telegeography.com/research-services/global-internet-geography/index.html
39. Ok, so what can be done ?
● Research on Downscaling the Semantic
Web will bring data sharing to everyone,
regardless of Internet connectivity
○ https://worldwidesemanticweb.wordpress.com/
● Creating mesh networks is a first step into
connecting devices
41. How to choose
● Reasons to prefer a mesh network
○ More robust
○ Faster to deploy
○ Example: Sensors network
● Reasons to prefer an infrastructure network
○ Better for the quality of service
○ Shared installation costs
○ Example: Universities' WiFi
● It is also possible to combine the two
42. Connecting remote areas
● Long wires
○ Use optic fiber and/or copper wires
● Long range Wireless
○ Use WiFi technology (IEEE 802.11b) to connect
relays over long distance
○ Network technologies 2G (GSM), 3G (WCDMA,
EDGE) and 4G (LTE, WiMax) are licenced
● Sneaker net
○ Use removable media to move information among
storage devices
48. What to pick?
● Keep in mind three things
○ The context of usage (dust, humidity, sunlight, ...)
○ The target users
○ The target use-cases
○ The resources needed
● The computer you have at home is most
probably not the one you would use in a
developing country
● There exist some optimized hardware
49. The Classmate from Intel
● Rugged
● Touch screen
● Handle
● Speakers
● Microphone
● Camera
50. The XO-1.75 and XO-4 from OLPC
● Use a couple of Watts
● User-defined sensors
● Accelerometer
● Light sensor
● Camera
● Rugged
● Speakers
● No movable part
● Microphone
51. SheevaPlug from GlobalScale
● ARM based CPU
● Ethernet, WiFi
● Solid state memory
● USB ports
● Here used as an e-
book server
52. T1 from Aleutia
● Sealed and fanless
● Operate under 45oC
● Atom processor
● SSD disk
53. A note on thin clients
● Thin client = very lightweight PC that
outsources all the computing to a bigger one
● Was among the first model of computing (e.
g. IBM's AS/400). Still popular in classrooms
settings thanks to its advantages
● Getting a new life with the increase of Cloud
computing (c.f. Google's ChromeBook)
55. Total cost of ownership (TCO)
● Total cost of acquisition and operation
● Computed over the expected operation time
Example: TCO of a car = car + parking place +
fuel + insurance + driving license + ...
For ICT: hardware + software + skilled staff +
infrastructure (electricity, network) + ...
56. TCO surprises
● Standalone data storage
solution (NAS) is often
less expensive than a PC
with software RAID
● Inkjet printers can be
more expensive than
laser printers
57. Things to keep in mind
● Think of the TCO when picking up software
○ Maintenance/Development is costly
○ Things that don't work now will be left aside
■ Remember the target users are not geeks!
● See how fit the software exactly is
○ Are vocal interfaces needed?
● Consider to which extent users can
appropriate and customise the software
58. Suggestions
● Prefer affordable and supported software
○ Individuals have few money but don't mind spending
it on things that matters
● Prefer open source software
○ Highest level of appropriation level
● Prefer software using open standards
○ If the software changes, the data can still be used
61. When you consider ICT in low
resources setting
● See what is already working and think how
ICT can be improve part of communication
● Think of all the side things
○ cost, environmental conditions, ...
● Design carefully your use-cases
● Employ appropriate hardware and software