> Any suggestions on how to tune TCP for this situation?
Roughly speaking, TCP throughput is window size / latency, and the default window size on Linux is usually pretty small, so for high latency connections (US to CN roundtrip is usually several hundred ms) the throughput would be bad. Increasing TCP buffer sizes goes a long way. In addition to buffer sizes, finer tuning is possible via other options/parameters; just find a guide online, there are plenty.
> Wouldn't the bandwidth cost of CDN be even higher than that of Alibaba Cloud?
Hmm, shouldn't CDNs for static assets be cheaper than bandwidth associated with compute instances in general? For instance, Alibaba Cloud's CDN offering starts at $0.04/GB (in Mainland China) for the first 50TB, and gradually decreases from there.[1]
Any suggestions on how to tune TCP for this situation?
> I suppose it's fine if you host static assets through a CDN and only use the expensive bandwidth from compute for truly dynamic content.
Wouldn't the bandwidth cost of CDN be even higher than that of Alibaba Cloud?