Leveraging the Mobile Web in Low-Resource Settings

Sam Sudar

1. ICTD
2. DUCES Model
3. Siskin

• ICTD
• DUCES Model
• Siskin

Information and

Communication

Technology for

Development

‘social, economic, and political development, with a particular emphasis on helping poor and marginalized people and communities’

https://en.wikipedia.org/wiki/Information_and_communication_technologies_for_development

Image Credit: JGI

In the Developing World:

​

• 80% have access to a mobile phone
• 31% have internet access

The World Bank. World Development Report 2016: Digital Dividends.

My Work

D.U.C.E.S.

D.U.C.E.S.

• ICTD
• DUCES Model
• Siskin

DUCES: helps people tweak requirements to fit technology

• Data collection
• Mobile workflows
• Where is complexity hiding?

​

Focus

Hope Study

0

3

6

• Desktop web app
• Log in, edit online
• Server sends email reminders
• Logic programmed on server

Seattle-based Study

• Must work offline
• Mobile-based edits
• Hired workers
• Shared mobile phones

Kenya-based Study

Hope Study

0

3

6

• Requires a developer
• They don’t have a developer

Problem

• Requirements can be flexible
• Don’t know what to change

​

​

DUCES insight

Data flow

User interface

Connectivity

Edit model

Server

Sources of Complexity

Data Flow

Unidirectional

Bidirectional

User Interface

Form-based

App-UI

Connectivity and Edit Model

Connected

Unordered

Disconnected

Transactional

• One-time creation
• e.g. logging temperature

• Logical edits
• e.g. bank withdrawals

Server

Bucket

Processed

Hope Study

0

3

6

UI

form-based

Data Flow

bidirectional

Connectivity

disconnected

Server

processed

Edit Model

unordered

Hope Study

0

3

6

Server

unprocessed

Data Flow

unidirectional

UI

form-based

Data Flow

bidirectional

Connectivity

disconnected

Server

processed

Edit Model

unordered

• Running for over 4 years
• Limited tech support
• DUCES helped identify complexity

​

​

Hope Study

My Work

• ICTD
• DUCES Model
• Siskin

Offline Educational Resources

Image Credit: Widernet

• $$$
• Can’t add content
• Hard to maintain

Not Perfect

• Free on existing infrastructure
• Local content curation
• Easy installation
• Minimal configuration

Ideally

Siskin lets you save and share web content directly from the browser.

Demo

In case the demo gods were against us...

Extension

App

Architecture

wikipedia.org/Cat

• Saves pages
• Annotates links
• Redirects to local resources

Extension

• Discovers peers
• Talks to peers
• List UI

App

• Find peers?
• Transfer data?
• Annotate Links?

How does this work?

How do we find peers?

Multicast DNS

• Network-local DNS

DNS-Based Service Discovery

• Use DNS as a database

Zero-conf: it just works

Review: Regular IP

172.28.7.80 Goes to one device

172.28.7.80

172.28.7.80

Review: Multicast IP

224.0.0.* Goes to all devices

224.0.0.*

224.0.0.*

Review: DNS

Response comes from remote server

A? www.google.com

A 216.58.216.164

Multicast DNS (mDNS)

Response comes from device on local network

A? www.google.com

A 216.58.216.164

224.0.0.251:5353

DNS-Based Service Discovery (DNS-SD)

Tyrion

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

PTR?

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

PTR

PTR

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

SRV?

Tyrion.siskin.tcp

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

SRV

tyrion.local 8888

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

A?

tyrion.local

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

A

172.28.7.90

172.28.7.90:8888

DNS-Based Service Discovery (DNS-SD)

Jon

Arya

Tyrion

A

172.28.7.90

172.28.7.90:8888

P2P Transfer

WebRTC

WebRTC

WebRTC

172.28.7.90:8888

WebRTC?

WebRTC.

Now we can talk, what do we say?

​

• List content
• Annotate links

List Content

List Content

List Content

khanacademy.org/math1

nytimes.com/story1

wikipedia.org/Cat

List Content

Annotate Links

Annotate Links: naively

khanacademy.org/math1

nytimes.com/story1

wikipedia.org/Cat

khanacademy.org/math1

nytimes.com/story1

wikipedia.org/Cat

…

…

…

…

…

​

1000 pages * 51 characters = 51 kB

10 peers * (51 * 9) = 4.59 MB

40 peers * (51 * 39) = 79.59 MB

Hard on bandwidth

List

• Is this page available?
• Yes, 100%!

Bloom Filter

• Is this page available?
• Probably. 99.9%

Bloom Filter

https://en.wikipedia.org/wiki/Bloom_filter

Bloom Filter

1,000 pages, 0.001 FP = 1.8 kB

10 peers * (1.8 * 9) = 162 kB

40 peers * (1.8 * 39) = 2.8 MB

Better on bandwidth

80 MB to 3 MB with 40 peers

Bloom Filter

Does this work?

Transfer Speed

Query Latency: 30 peers; 1000 pages

Bandwidth Impact: 1,000 pages per peer

Conclusion

• Free
• Can curate content
• Piggybacks on distribution
• Rachel and eGranary could use this approach as well

Richard Anderson

Matt Welsh

​

Clarice, Mitch, and Jeff

Saloni Parikh and the undergrads

Thank you!

Questions?

Paul G. Allen School for Computer Science and Engineering