Why are my socks proxies slow

Question

I have a linux vps, and i have tried a few socks proxy setups to test their performance: All tests were using speedtest.net

1. The standard ssh tunnel proxy 0.8mbit/s download and 0.1-0.2mbit/s upload speeds

2. dante-server proxy 1.3mbit/s download and 0.4-0.5mbit/s upload

I am wondering why are these speeds so slow? Is anything shaping them? Is it just the nature of socks proxies? I know that the ssh tunnel has to do encryption and what not so that is why its slow, but i was surprised to see that the second setup was also quite slow.

On the VPS i have received download speeds of 25MB/s per second (thats about 200mbit/s and upload speed of atleast 5MB/s (haven't got a good enough pipe to test anything faster).

The other option i was going to try is to setup OpenVPN and see how that goes, however i need to find a good tutorial as it's fairly complicated to setup.

So why is it so slow? How can i test to see where the bottleneck is? How can i make it faster >:D

Answer 1

A ssh SOCKS proxy isn't inherently slow - I can get throughput performance similar to a direct Internet connection (latency is of course impacted the most). Given the direct speeds listed, I suspect it's not the VPS provider (they could be restricting SSH bandwidth, but that type of meddling seems unlikely from a real provider). Perhaps your local network is either prioritizing web/short-duration connections and has congestion causing non-web/long-duration to be restricted, or is specifically limiting ssh/socks? If the traffic shaping is port-based rather than DPI, you may be able to have ssh listen on port 443 where other than the first few bytes it looks somewhat like HTTPS. If the shaping is based on connection duration, you may get better results from testing immediately after connection.

Answer 2

I have the same confusion as you do. But on my side, dante-server (400KB/s down and up) is much slower than a ssh tunnel proxy (2MB/s down and up, nearly equal to direct link speend between client and server with random ports).

I have also tested random ports and ports like 8080 and 1080 without noticing a difference.

Mysterious, ha?

Answer 3

This is an inherit problem of OpenSSH: Window scaling in TCP. If you want to get around that, check out High Performance SSH. But beware, some cipher modes HPN-SSH adds to OpenSSH are not so secure... e.g "none" ;-)

http://www.psc.edu/networking/projects/hpn-ssh/

EDIT: The site seems to be down from here. Hope you can reach it. If not, just use your favorite search engine :)