Accessing Google Scholar under Extreme Internet Censorship A

Accessing Google Scholar under Extreme Internet Censorship: A Legal Avenue Zhen Lu, VMware Zhenhua Li, Jian Yang, Tianyin Xu, Ennan Zhai, Yao Liu, Christo Wilson, Tsinghua University / Tsinghua University UCSD Yale University Binghamton University Northeastern University luzhen 02@gmail. com Dec. 15, 2017

Outline 1 2 3 4 Background Feasibility Study Scholar. Cloud System Performance 5 Summary

Internet censorship across the world China, Iran, Russia, Thailand … q Political websites are blocked mostly. q Non-Political websites may be blocked incidentally!

Troubles of scholars Find a topic -> Google Scholar search -> Read papers Google Scholar index includes roughly 160 million documents

Troubles of scholars Find a topic -> Google Scholar search -> Read papers Chinese researchers CANNOT access it.

Ways to bypass censorship All roads lead to Rome! VPN Tor Shadowsocks Other Proxies

Problems of circumvention system Difficult to use not available Expensive Illegal Unstable/ Why not build a system for

2 Feasibility Study 8

Internet censorship in China Technical part Governmental part Get registered on MIIT website -> Provide public Internet service legally

Mainstream Choice 371 faculty members and students from THU q VPN is the most popular way q Shadowsocks also plays an important role

Comparison q VPN √ Stable × Exposed to GFW (DPI) q open. VPN √ Secure × Bad Usability

Comparison q Tor × Exposed to GFW (DPI) √ Automatic Configuration q Shadowsocks √ Easy to setup √ Traffic split

3 Scholar. Cloud System 13

Combine advantages together √ Stable √ Secure √ Auto-config √ Traffic split Internet Censorship

Combine advantages together √ Stable √ Secure √ Auto-config √ Traffic split Internet Censorship http: //scholar. thucloud. com

Combine advantages together √ Stable √ Secure √ Auto-config √ Traffic split Internet Censorship Launched in Jan. 2016 2000+ users/ 700+ daily active users 2. 2 USD operation cost daily http: //scholar. thucloud. com

Key points Split-proxy and auto-config http: //pac. scholar. thucloud. com/df 70 d 12 a-f 586 -11 e 5 -980 a-08002705856 d. pac Proxy Automatic Configuration file (PAC) Internet Censorship Web proxy

Key points Split-proxy and auto-config http: //pac. scholar. thucloud. com/df 70 d 12 a-f 586 -11 e 5 -980 a-08002705856 d. pac Proxy Automatic Configuration file (PAC) Web proxy

Key points Message blinding Public well-known encryption schemes suffer from DPI. (Liang Wang, CCS 2015) Internet Censorship

Key points Service legalization ICP Reg. #15063437 in China Internet Censorship

4 Performance 21

Network experience 1 TCP Stream S. S. only. 10 s timeout. Main stream First-time* access cost

Network experience 2 Page load time u Worst first-time latency 15 s —— Tor u Wost subsequent latency 4. 5 s —— S. S. First time access suffers great latency: DNS resolution / Web cache / Google record*

Network experience 3 Round-Trip Time u Tor has the largest average RTT u Tor is also the most unstable one Shadowsocks shows the same average TTL but much larger max TTL* Authentication

Network experience 4 Packet Lost Rate u Tor suffers from 4. 4% PLR u Compared with normal website access (Amazon） 0. 2% S. S. is much worse than Scholar. Cloud again! DPI (Message Blinding)

Client side overhead Network Traffic CPU Memory Differences are not remarkable, while Scholar. Cloud is no worse than the average.

Scalability u. Shadowsocks PLT sharply grows after 60 u. Scholar. Cloud acts well in large concurrency

Summary p. Legally accessing Google Scholar under extreme censorship p. Design and deploy Scholar. Cloud to provide web proxy service to thousands of users, with high performance, usability, robustness against censorship, and low overhead p. Measure and analyze five main circumvention system on network performance, client side overhead, robustness against censorship and scalability. We provide a new perspective to Internet censorship in China. We hope that our experiences will help the community better understand China’s Internet censorship ecosystem.

The End Thank you!

System Architecture p Portal website system p Split-proxy system p User control system

System Architecture