Shadowsocks imposes an extra per-session TCP connection for user/password authentication plus a 10-second keep-alive timeout, causing an average page load time of 3.7 seconds and a sharp PLT inflection when concurrent clients exceed 60. In contrast, ScholarCloud (split-proxy, no per-session auth handshake) achieves 1.3 seconds average PLT with linear scalability up to 180 concurrent clients. Native VPN and OpenVPN also scale linearly; Shadowsocks is the only tested solution with a non-linear degradation point.
From 2017-lu-accessing — Accessing Google Scholar under Extreme Internet Censorship: A Legal Avenue
· §4.3
· 2017
· Middleware
Implications
Per-session authentication round-trips (as in default Shadowsocks configuration) are a significant PLT bottleneck — circumvention proxies should use session-resumption or connection-pooling to amortize handshake cost.
The Shadowsocks scalability cliff at 60 concurrent clients suggests that shared proxy server deployments should account for per-session state overhead when capacity-planning for real user loads.