UNDERSTANDING INTERNET USAGE AND NETWORK LOCALITY IN A RURAL - PowerPoint PPT Presentation

UNDERSTANDING INTERNET USAGE AND NETWORK LOCALITY IN A RURAL COMMUNITY WIRELESS MESH NETWORK Adisorn Lertsinsrubtavee, Liang Wang, Nunthaphat Weshsuwannarugs, Arjuna Sathiaseelan, Apinun Tunpan, Kanchana Kanchanasut, Jon Crowcroft Computer Laboratory, University of Cambridge intERLab, Asian Institute of Technology 1

OUTLINE ➤ Internet in Rural Area of Thailand ➤ Community Network ➤ TakNet CWMN ➤ Social Interview ➤ Traffic Measurement & Data Analysis ➤ Discussion and Takeaways 2

INTERNET IN RURAL AREA OF THAILAND ITU -D Thailand’s Internet 28.92% Penetration ranked 6th in AEC, as of 2014 Low demand of High cost for using Internet infrastructure investment It is not cost-effective for ISP to invest network infrastructure Digital Divide 3 Thai National Statistical Office, http://web.nso.go.th/

COMMUNITY NETWORK Wireless ad-hoc network sharing the Internet gateway Successful Community Networks ! Internet 4

TAKNET CWMN Thai Samakhee a small rural village in northern Thailand 50 households with 300 population Before 2013 TakNet CWMN 2 ADSL links provided by ISP I nternet cost is shard among villagers 28$ /month for a subscription 5$ /month for a subscription A ttract villagers to use the Internet 5

TAKNET CWMN Core Router R4 Ad hoc link Access Router OLSR ADSL gateway R7 R12 R3 R14 R5 R13 Core R11 Router WiFi ADSL R2 Gateway 16 GB external R1 R8 storage R10 R9 R6 14 access routers (TPlink MR 3040) 1 core router (Unifi UAP) OpenWrt, Attitude Adjustment 12.04 6

UNDERSTANDING THE INTERNET USAGE Traffic Social Interview Measurement ➤ Lightweight measurement ➤ Well-defined questionnaire ➤ Traffic volume ➤ Personal information ➤ ifconfig - 60 sec interval ➤ Typical usage ➤ Internet usage ➤ User feedback ➤ tcpdump - HTTP request ➤ 30 mins interview ➤ Filter out the URL ➤ Free-style conversation 7

VS SOCIAL INTERVIEW User information Device used 60% 40% 18 34 52 interviewees 91% Kids Teens Adults Popular content over 22 16-21 8-16 Monthly 8 Adults wages 14 30 Teens Kids 140$ - 560$ 8

7.6% 12.5% 06:00 - 12.00 12:00 - 17.00 22:00 - 06.00 87% 71% 30% 21% SOCIAL INTERVIEW Social Communications Usage pattern 33% 81% of Line users have local contacts within the same village 10-20% of messages 80% 17:00 - 22.00 exchanged among local users User feedback 85% of users install CM battery application (expect to improve their WiFi speed) 4 hours per day 4 users opted out due to the extra cost incurred by on Internet usage electricity bill (just 1-2$/month) 9

TRAFFIC USAGE Upload Download 1500 14000 Feb Feb Mar Mar 12000 Apr Apr 10000 Traffic load (MB/day) Traffic load (MB/day) 1000 8000 6000 500 4000 2000 0 0 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 Day (1 Febuary 2015 − 30 April 2015) Day (1 Febuary 2015 − 30 April 2015) Traffic Growth Feb - Mar 15 Mar - Apr 15 Upload 25% 20% Download 28.9% 15% 10

TRAFFIC USAGE PATTERN Traffic almost reached the committed speed (4Mbps) Time of day (Hours) 3.5 Mon Tue Wed Thu Fri Sat Sun 3 Traffic load (Mbps) 2.5 2 1.5 1 0.5 0 01.00 07.00 13.00 19.00 01.00 07.00 13.00 19.00 01.00 07.00 13.00 19.00 01.00 07.00 13.00 19.00 01.00 07.00 13.00 19.00 01.00 07.00 13.00 19.00 01.00 07.00 13.00 19.00 Dual off-peak hours 13:00 - 16:00 How can we efficiently utilise these 22:00 - 07:00 resources? 11

CONTENT POPULARITY 14 Zipf distribution with alpha = 0.57 12 Fraction of requests (%) 10 8 6 4 2 0 ksmobile naver baidu google 3g.cn umeng instagram fb youtube animaljam Domain (url) ➤ Commonly known alpha (0.9 - 1.1) ➤ Mixture of misbehaviour domains ➤ Some valuable domains such as education and local newspaper are pushed to the tail 12

CONTENT POPULARITY 14 Zipf distribution with alpha 1.05 12 Fraction of requests (%) 10 8 Education 6 website appears 4 2 0 naver google instagram fb youtube apple msn ytimg thscore dek − d Domain (url) Remove all suspicious domains 13

ANALYSIS OF SUSPICIOUS DOMAINS Jain’s fairness index The higher value indicates the requests are more uniformly distributed 0.8 0.7 Jain Fairness Index 0.6 0.5 0.4 0.3 0.2 0.1 0 ksmobile naver baidu google 3g instagram umeng fb youtube animaljam Large amount of requests were generated by some specific users 14

APPLICATION BEHAVIOUR Inter arrival time of suspicious domain Almost 80% of requests are 1400 1200 made with an inter arrival Number of requests less than 2s 1000 800 600 These requests were generated by the baidu 400 browsers 200 0 0 5 10 15 20 25 Interval (s) baidu.com 15

MISINFORMED KNOWLEDGE ➤ Several users misuse an application ➤ Considering ksmobile domain ➤ Android application —> CM Battery ➤ To save the battery power ➤ But! the villagers believe that it can use to accelerate the WiFi speed ➤ Fact! it generates a lot of request to ksmobile domain ➤ Observation! advertisements are automatically downloaded to users’ mobile phone ➤ 85% of villagers use this application 16

LOCALISED COMMUNICATION HTTP request to Line server from each router 14 13 12 Line application (naver) 11 10 9 Rotuer ID 8 7 Line 6 server 5 4 3 Notification 2 Send 1 0 2 4 6 8 msg Time of day (second) 4 Retrieve x 10 msg Sender Receiver 17

LOCALISED COMMUNICATION Achieve10% - 15% of 14 Varying window size 1s - 5s identified pairs 13 12 11 10 From interview, 10% - 20% 9 were sent to the local 8 7 contacts 6 5 4 3 2 1 3.2 3.205 3.21 3.215 3.22 3.225 3.23 4 x 10 A pair of communication represents the localised communication 18

TAKEAWAYS What are the impacts of TakNet ? ➤ TakNet is able to create a demand within the community for Internet access ➤ Number of Internet users in TakNet is increased significantly ➤ Villagers gain significant benefits form the Internet ➤ TakNet is a catalyst for changes: ISPs expand more backhaul to cover the villages 19

TAKEAWAYS Is there a universal model for all rural settings? ➤ Traffic pattern ➤ Asia 1 - TakNet: Dual-peak pattern ➤ Africa 2 and Europe 3 : Single peak: ➤ Localised communication ➤ TakNet: 10 - 15%, Africa 2 : ~50% ➤ Social Communications ➤ OSN (e.g., FB, Twitter) and email are popular services in rural Africa 2 . ➤ Instant messaging is the most dominant service in TakNet 1 B. Du, et al. Analysis of www traffic in cambodia and ghana. In WWW ’06. ACM, 2006. 2 D. L. Johnson, et al. Network traffic locality in a rural african village. In ICTD. ACM, 2012. 20 3 A. Sathiaseelan, et al. A feasibility study of an in-the-wild experimental public access wifi network. In ACMDEV , 2014

TAKEAWAYS What are the potential solutions to improve TakNet? ➤ The available 4 Mbps bandwidth may be saturated soon in the near future. ➤ Can we simply expand the the link capacity or add more gateway? ➤ Villagers are very sensitive to the cost ➤ Can we utilise the off-peak hours with content/service caching ? ➤ Identify the true valuable contents ➤ Efficiently remove the suspicious domains ➤ New technologies ➤ Information Centric Network ➤ Service migration - virtualisation, container 21

THANK YOU Q&A Adisorn Lertsinsrubtavee al773@cam.ac.uk 22