N ETWORK T RAFFIC L OCALITY IN A R URAL A FRICAN V ILLAGE David L. Johnson, Elizabeth M. Belding and Gertjan van Stam (davidj, ebelding)@cs.ucsb.edu,
[email protected] P ROBLEM
FACEBOOK INSTANT MESSAGE INTERACTION GRAPH
S OCIAL GRAPH STATISTICS
Traffic usage patterns in rural networks are not well understood. Our analysis reveals an abundance of localized traffic passing through servers located in the West due to the webcentric model of Internet usage. The centralized architecture of the Internet, using a few monolithic services, is inefficient in rural networks that utilize slow satellite gateways to the Internet.
The clustering coefficient measures the tendency of nodes to cluster together; a higher value represents more localized cliques. • 54% of instant messages were between local users; 35% of users were local (7.5% of local users travelled). • Average node degree between local users is 3.6; from local to external users it is 5.3. • Average number of messages between local users is 93 and from local to external users it is 76. • Clustering coefficient for instant messaging is 0.1 (Facebook average is 0.164 for friend lists). Strongly connected local community as regular interactivity occurs with a fourth of a user’s Facebook friend list.
T RAFFIC COLLECTION
P OTENTIAL L OCAL TRAFFIC Skype and Bittorrent
35.10%
Google mail via imap
22.74%
Facebook
8.22%
Web Hosting
• Small set of key users have strong links to the outside world (fan motifs) • Some isolated communities • Strongest bonds between local users • External users usually have one or two local IM contacts (gatekeepers)
Users represented by nodes and edges denote conversations; thicker edges indicate that more messages were sent between users.
FACEBOOK TRIANGULATION
I NTERNET U SAGE 15.76%
*facebook.com
1 0.8 0.6 0.4 0.2 0
Local to local Local to external
0
0.24% 8.88% 7.88%
*postzambia.com
4.45% 4.30%
*yahoo.com
3.91% 4.65% 2.54% 1.63%
lusakatimes.com
2.05% 1.56%
doubleclick.net
1.93% 1.01%
2011 (Feb−Mar) 2010 (Feb−Mar)
Social networking (Facebook and Twitter) is three times more popular than web search.
5
10
15
20
25
30
35
Social Degree
11.47%
*google.com
*windowsupdate.com
20.26%
Fraction of Users
*twitter.com
S OCIAL GRAPH ANALYSIS Fraction of Users
Captured 2 months of traffic from the wireless network in Macha, Zambia, serving 300 users over a 256kbps satellite link. All traffic headers were captured at the gateway switch.
• 573 unique users (200 local users) • 14,217 unique instant messages sent between 726 unique user pairs over 2 months
Same IM message sent to both sender and receiver and displayed on both IM web clients. Local conversations receive packets with same user pair on two different local machines.
1 0.8 0.6 0.4 0.2 0
Local to local Local to external
7.05%
Web mail
3.06%
Podcasting
3.02%
File sharing
Upload traffic > 100K
1.60%
Analysis of outgoing traffic to sites that could facilitate local file sharing: • 8.1% of outgoing traffic would be saved assuming similar Facebook locality. • Our solution, VillageShare (ACMDEV’12), provides web-centric localization to save satellite bandwidth. • No evidence of any direct local file sharing in the village.
O NLINE SURVEY Online survey of random selection of 77 users in Macha (25% of user base): 20 to 30 yrs old
0
100 200 300 400 500 600 700 800 900 1000
Number of messages sent
Strong locality of interest between small cliques of local users. Larger node degree to external users with weak messaging.
Female
69% 34%
Use Internet > 3 hrs/day Internet at home
67% 49%
Use social networking Use Instant messaging
91% 72%
