Tag: tcp

I am currently dealing with a network problem on high latency internet links (100-400 ms). I ran a Minecraft network, and recently moved it to a separate data center to get a server with a better CPU and more RAM. The users of this server are spread all over the world. Before the change, the server was in Montreal and users in Europe had a latency of ~ 100-200ms, and those in Australia had a latency of ~ 200-300ms. After the change, the server is in Germany, users in North America get a latency of ~ 100-200 ms, and users in South America and Australia get a latency of 200-400 ms. Overall, the latency is pretty similar, but who gets excellent latency and who gets tolerable latency varies (note that Minecraft isn't very latency-sensitive overall, especially compared to most video games). There is also no significant packet loss, as measured by the MTR and ping tools. Also, the software on both servers is almost identical. Both servers are running Debian 10, and I checked all the software + settings that are not in the APT repositories to send, while reinstalling the exact same packages via apt. As such, the software configuration must be essentially identical.

However, many users have connection problems. It appears to only occur around 6:00 PM (± a few hours) in the eastern United States. The connection problem specifically takes the nature of the performance of all TCP connections to be appallingly low. With an SSH + SOCKS proxy, it took minutes Loading a normal web page (Gmail), and in-game, it often takes minutes even a simple chat message to arrive if a few MB of global data is transferred. The effective latency of a TCP connection (for example, the time it takes for a chat message to pass through it) increases unreasonably massively when data is placed on that TCP connection. A normal SSH session with only one terminal is basically fine, and the game is fine overall if not much is happening, but as soon as something of significant size is sent over TCP and is during the time mentioned above then performance it breaks and the latency over TCP (but not through ping) becomes irrational, even several minutes in the worst cases. When this problem first occurred, there was a significant packet loss (~ 25%) which I thought was to blame, but that packet loss no longer occurs (according to ping etc.) but the problem persists. The packet loss, but not the troublesome symptoms, disappeared after I made a general report to the new host about the packet loss, but before I could give them more detailed data with MTR as requested in their response to that report. My impression of the host is that nothing was changed, but who really knows.

As such, at this point, I suspect that the relevant difference between servers is that the previous host (OVH) performs some kind of tweaking of their operating system images (something I know to be the case), and that the new host ( Hetnzer) no.

I suspect this setting has something to do with the size of the TCP window, but when I tried to manipulate those settings to make changes, the settings didn't seem to do what they were supposed to. Specifically, when I configure the various net.ipv4.mem or net.core.mem settings i find listed on the internet via sysctl, the size of the window that iperf select (or the maximum that is allowed to select, when using the -w option) seems to take a random value with no discernible relationship to me with the values ​​I configured via sysctl, rather than behaving the way I expected, where its maximum value is simply what I configured via sysctl. Note that iperf -s It misbehaves even before a client connects to it, so not making the same changes on the client is not a plausible explanation.

As such, I wonder 2 things:

1) How can I repair my server and allow the latency on TCP connections to be similar to the actual latency on the link, even at peak times and under moderate load (a few mbps)?

2) How can I reliably and predictably change the TCP window size of all applications? (Or equivalently, what's going on with the sysctl setting being applied in seemingly random ways? What's the pattern I'm missing?)