1 of 48

A Framework for Improving

Web Affordability and Inclusiveness

Rumaisa Habib*

Sarah

Tanveer*

Aimen

Inam

Haseeb

Ahmed

Ayesha

Ali

Zartash

Uzmi

Zafar

Qazi

Ihsan

Qazi

* Joint first authors

1

Rumaisa Habib Stanford University

2 of 48

How ‘worldwide’ is access to the World Wide Web?

How would stakeholders benefit from Web equity?

How do we work towards equitable Web access?

2

Rumaisa Habib Stanford University

3 of 48

3

Social stigma

Coverage

Censorship

Digital Literacy

Rumaisa Habib Stanford University

4 of 48

4

Social stigma

Coverage

Censorship

Digital Literacy

Cost of web access

Rumaisa Habib Stanford University

5 of 48

5

Broadband price

Webpage size

Cost of web access

Rumaisa Habib Stanford University

6 of 48

Difficulties in web access

  • 48% of respondents had difficulty paying for their mobile data
    • World Bank survey in 11 emerging countries
  • Rising webpage sizes and stagnant broadband prices
    • 13.8x median increase in page size in the past decade
    • 2020-21
      • 0.02% decrease in broadband price
      • 2.3% increase in webpage size
  • Each access has become more expensive

6

Rumaisa Habib Stanford University

7 of 48

Variations in…

206 countries

ITU dataset

99 countries

Alexa top 1000

~72k webpages

7

broadband price

web complexity

Rumaisa Habib Stanford University

8 of 48

Variations in…

Three data plans

94 countries do not meet the 2% target

(set by the UN Broadband commission)

for the 2GB DO plan

206 countries

ITU dataset

99 countries

Alexa top 1000

~72k webpages

8

}

web complexity

broadband price

Rumaisa Habib Stanford University

9 of 48

Variations in…

non cached Page sizes in…

…developing regions: 2.87 MB

…developed regions: 2.64 MB

206 countries

ITU dataset

99 countries

Alexa top 1000

~72k webpages

9

web complexity

broadband price

Rumaisa Habib Stanford University

10 of 48

PAW Index

novel fairness metric to quantify affordability…

reduction required in avg. page size by each country to equalize web access

10

Rumaisa Habib Stanford University

11 of 48

PAW Index

reduction required in avg. page size by each country to equalize web access

11

Rumaisa Habib Stanford University

12 of 48

PAW Index

reduction required in avg. page size by each country to equalize web access

12

48* out of 96 countries have a PAW > 1

(for at least one plan)

* all of which are developing

Rumaisa Habib Stanford University

13 of 48

Approaches

  • Reducing broadband prices
    • ISPs have limited margins already
    • Large systematic change that cannot be easily done

  • Reducing webpage sizes
    • Can be fine tuned
    • Easy to implement

13

Rumaisa Habib Stanford University

14 of 48

Approaches

  • Reducing broadband prices
    • ISPs have limited margins already
    • Large systematic change that cannot be easily done

  • Reducing webpage sizes
    • Can be fine tuned
    • Easy to implement

14

Rumaisa Habib Stanford University

15 of 48

How ‘worldwide’ is access to the World Wide Web?

How would stakeholders benefit from Web equity?

How do we work towards equitable Web access?

15

Rumaisa Habib Stanford University

16 of 48

Do users even want their webpages reduced?

16

Quality

Accesses

Quality

Accesses

🤔

Rumaisa Habib Stanford University

17 of 48

Do users even want their webpages reduced?

  • 100 participants
  • 10 webpages

with varying levels of reduction

17

Rumaisa Habib Stanford University

18 of 48

Do users even want their webpages reduced?

  • 100 participants
  • 10 webpages

with varying levels of reduction

18

willing to trade off webpage quality for number of accesses

Rumaisa Habib Stanford University

19 of 48

19

Rumaisa Habib Stanford University

20 of 48

20

Rumaisa Habib Stanford University

21 of 48

Other Stakeholders

21

Website Operators

Mobile Network Operators

more users online increased revenue

Rumaisa Habib Stanford University

22 of 48

So now that we know that…

… this problem exists

22

Rumaisa Habib Stanford University

23 of 48

So now that we know that…

… this problem exists

… people are willing to adopt solutions

23

Rumaisa Habib Stanford University

24 of 48

How ‘worldwide’ is access to the World Wide Web?

How would stakeholders benefit from Web equity?

How do we work towards equitable Web access?

24

Rumaisa Habib Stanford University

25 of 48

AW4A

Affordable Web For All

25

Rumaisa Habib Stanford University

26 of 48

Prior work

What went wrong?

26

Rumaisa Habib Stanford University

27 of 48

Prior work

What went wrong?

1 Breaks pages

2 Lack of web developer consent (impacts revenue)

3 Violated net neutrality principles

4 Broke end-to-end encryption

27

Google Weblight 1,2,4

Facebook Free Basics 3

Opera Mini 1,2

Brave 1,2

Rumaisa Habib Stanford University

28 of 48

Goals & principles

28

Rumaisa Habib Stanford University

29 of 48

Goals & principles

29

Use PAW as a target

G1

sizes based on affordability

Rumaisa Habib Stanford University

30 of 48

Goals & principles

30

Use PAW as a target

Server-side webpage transcoding

G1

sizes based on affordability

G2

user privacy & developer consent

Rumaisa Habib Stanford University

31 of 48

Optimization problem

  1. Maximize page quality
  2. Given a page size limit

31

page quality

sum of object sizes

Rumaisa Habib Stanford University

32 of 48

Optimization problem

  • Maximize page quality
  • Given a page size limit

32

Rumaisa Habib Stanford University

33 of 48

Two stages

33

Rumaisa Habib Stanford University

34 of 48

Two stages

34

Rumaisa Habib Stanford University

35 of 48

Two stages

35

Rumaisa Habib Stanford University

36 of 48

Two stages

  • Grid search vs HBS (two columns)

36

Rumaisa Habib Stanford University

37 of 48

Zooming into stage 2…

37

HBS

  • Heuristics Based Approach
  • Uses a set of heuristics
  • Lower quality pages
  • Linear run time

Grid Search

  • Brute Force Approach
  • Searches discrete space
  • Higher quality pages
  • Exponential run time

VS

Rumaisa Habib Stanford University

38 of 48

A sneak peak into the (image) heuristics

38

Area on the webpage

Bytes efficiency

Image A

Image B

Greater Δbytes

Lesser Δbytes

for the same change in quality:

Rumaisa Habib Stanford University

39 of 48

A sneak peak into the (image) heuristics

39

Area on the webpage

Bytes efficiency

Image A

Image B

Greater Δbytes

Lesser Δbytes

for the same change in quality:

Rumaisa Habib Stanford University

40 of 48

A sneak peak into the (image) heuristics

40

Area on the webpage

Bytes efficiency

Image A

Image B

Greater Δbytes

Lesser Δbytes

for the same change in quality:

Rumaisa Habib Stanford University

41 of 48

Quality of webpages produced by HBS

Half of the pages maintain a quality* of 0.98 or higher

* weighted avg. of the quality of objects on a page

41

Rumaisa Habib Stanford University

42 of 48

Quality of webpages produced by HBS

Users slightly preferred HBS over

42

Brave Opera Mini

Rumaisa Habib Stanford University

43 of 48

Future work… what now?

  • Involve governments to optimize regional sites
  • Make the framework more accessible to developers
    • Deploying it and involving it in development pipelines
  • Video affordability
    • How can we transcode videos?

43

Rumaisa Habib Stanford University

44 of 48

Conclusion

We show that…

… lack of affordability is a key barrier in accessing the web

… users are willing to trade quality for quantity of access

… there are practical solutions to improve affordability

44

Rumaisa Habib Stanford University

45 of 48

Thank you!

For more details…

contact 📨 us at rumaisa@stanford.edu or 23100099@lums.edu.pk

… view our source code 💻 at https://github.com/nsgLUMS/sigcomm2023-aw4a

… and check out the full paper 😎

45

Rumaisa Habib Stanford University

46 of 48

Backup Slides >>

46

Rumaisa Habib Stanford University

47 of 48

Gzip compression

47

Rumaisa Habib Stanford University

48 of 48

PAW Index equation

𝑃𝑖 : average broadband price in region 𝑖

𝑃𝑇 : target broadband price (2% of a country’s GNIpc)

𝑊𝑖,𝑎𝑣𝑔 : average page sizes in region 𝑖

𝑊𝑔𝑙𝑜𝑏𝑎𝑙 : average page size globally

48

Rumaisa Habib Stanford University