1. PPWEB: A Peer-to-Peer Approach for Web Surfing On the Go
Ling-Jyh Chen, Ting-Kai Huang
Institute of Information Science, Academia Sinica, Taiwan
Guang Yang
Nokia Research Center, Palo Alto, US

2. Motivation Web surfing is part of our life.
How can we surf the Web when we cannot directly access the web pages?
No connections
Censorship
Mobile devices are hugely popular.
How can we browse the Web when we are on the go?
Cellular
Wi-Fi Hotspots

3. Previous Solutions Offline-based approaches Cache-based approaches
Gnu Wget
Wwwoffle
Well-known web browsers
Cache-based approaches
Push based (Aalto ‘04, Costa-Montenegro ‘02, Spangler ‘97)
Pull based (Jiang ’98a, Jiang ’98b, Padmanabhan ‘96)
Infostation-based approaches
Mobile Hotspots (Ho ‘04)
Thedu (Balasubramanian ‘07)
offline-based的方法，主要是事先將網頁下載至本地的存儲空間中
暫存器輔助類的方法則會去去猜測使用者可能會想要瀏覽的網頁，並且由程式本身主動從網路上下載回來
業者推播(push mode)模式，如[6] [13] [23]，或者是
程式主動索取
固定式資訊站為基礎類的方法是利用架設好的資訊站來扮演網際網路和行動網路(mobile network)中間的橋樑

4. Previous Solutions (Drawbacks)
Offline-based
manually download web documents
limited number of web pages
Cache-based
Tremendous storage overhead
You still need a data plan to surf.
Infostation-based
Dedicated Infostations needed
Single point of failure

5. Assumptions We Make All peers collaborate.
All peers have local connectivity
WiFi, Bluetooth, etc.
All peers are mobile.
Some peers have Internet access.

6. What We Propose: Scenario 1
Internet
HTTP
Gateway Peer:
A peer who can access the Internet directly

7. What We Propose: Scenario 2a
Gateway Peer (B)
Vanilla Peer (A):
Peer that cannot access Internet directly

8. What We Propose: Scenario 2b
Vanilla Peer (A)
Vanilla Peer (B)

9. B gets A’s request
B is a GP Y N
B and A are
connected
B has the requested
web content Y N Y N
Direct forwarding
The request has been
relayed H times
B and A are
connected N N
Collaborative
forwarding Y Y
Indirect
Forwarding
Do nothing
Request
Forwarding

10. Direct Forwarding vs. Indirect Forwarding
B has complete content =>Direct Forwarding algorithm
B may only have partial content =>Indirect Forwarding algorithm
Further passing the request message using Request Forwarding algorithm

11. Cooperative Forwarding Algorithm
Increase the packet delivery ratio and decrease the request response time
HEC-PF
Hybrid Erasure Coding Algorithm (H-EC)
Probabilistic Forwarding Algorithm
Erasure codes increase error tolerance.
Extra caching increases hit ratio in the future (esp. for popular pages).

12. Evaluations
Evaluate the performance of PPWEB scheme against Mobile Hotspots scheme
Service ratio and traffic overhead
DTNSIM: Java-based simulator
Real wireless traces
UCSD (campus trace)
iMote (Infocom ‘05)

13. The Properties of two network scenarios
Trace Name
iMote
UCSD
Device
PDA
Network Type
Bluetooth
WiFi
Duration (days) 3 77
Devices Participating
274
273
Number of Contacts
28,217
195,364
Avg # Contacts/pair/day

14. Parameter Settings Number of VPs: Number of requests:
20% of the other peers
Number of requests:
first 10% of simulation time with a Poisson rate of 1800 sec/request.
The HTTP requests:
top 500 requested web pages,
campus proxy server of NTU, Apr.-Sept
Traffic:
iMote 15 request 44 peers
UCSD 370request 44 peers

15. Scenario 1: UCSD
γ= 20%
γ= 60%
iMote之所以會讓improve變大，是因為imote的internal node 和external node的差別。
Ucsd的每個node的contact次數類次的話，就不會有這樣的問題。

16. Scenario 2: iMote γ= 20% γ= 60% 坐標的說明 一波一波是因為會議的作息
R愈大，service ratio隨之成長

17. Traffic Overhead γ PPWeb Mobile Hotspots Normalized Overhead iMote 20%
4,298
1,423
3.02
40%
3,986
1,591
2.50
60%
3,938
1,694
2.15
UCSD
211,604
65,723
3.22
189,702
64,858
2.92
163,883
62,663
2.62
Replication factor of erasure coding = 2
Aggressive forwarding phase of the HEC-PF: make one more copy
The upper bound of the traffic overhead : 2*2=4

18. Summary PPWEB is a peer-to-peer solution to enable mobile web surfing.
No constant Internet access is required.
No dedicated servers are required.
It implements a Collaborative Forwarding algorithm that takes advantage of opportunistic encounters.

19. Thank You! 20