Author: tavimori

I haven’t thought a lot, but the following two are correct:

1. being systematic is optimal if it appears only once.
2. delivering a systematic copy is still optimal if a prefect feedback reports original losing.

Then what is the cost for other situation?

During the epidemic of 2019-nCoV, my supervisor shares a thought to me. He thinks the network communication should always be characterized as two problems: 1. characterizing the end-to-end network channel. 2. optimize channel coding for particular use.

I somehow believe in the idea, but I’d appreciate if we can give a theoretical proof.

Of course, the article won’t discuss this. Instead, I tried to use this to optimize the campus network for me.

Campus Network Problem

The campus network provides a good channel to the internet, in a sense:

1. I can acquire multiple IP, with independent bandwidth limit.
2. Good QoS, which can be reflect as fewer jitters and stable bandwidth.
3. Multiple ISPs (though I don’t have control of routing)
4. UDP is good as well

But also problems:

1. Slow (or not that fast?) – limited to 50Mbps per IP, but has 1Gbps Ethernet
2. CHINA intranet only
3. BAD routing policy

All in all, we choose to:

1. use third-party tunneling to pass the national firewall.
2. use client to server end-to-end MP-TCP aggregation to improve the bandwidth.

[mobile phones, PCs and tablets]

[our local gateway]
multiple IPs, each as (L2TP) client over (UDP) over (encrypted socks5 proxy)

[third-party cloud]
(UDP) tunneling as bridge between our local gateway and cloud, provided by (encrypted socks5 proxy)

[our cloud gateway]
(L2TP) server, (encrypted socks5 proxy) server

[REAL AND BETTER INTERNET]

[TO BE CONTINUED]

I will note others in the future, and in fact, I’m considering a TCP over TCP and perhaps a real NC-TCP.