User X on forum Y has made an observation that on his system Tixati is limiting download speed for each peer connection over TCP for no reason at all. Any single peer rarely reaches 85 KB/s, and is usually at 60 KB/s. RQ values are strictly 9/9 or 6/6, respectively, and they never grow. If UDP connections are enabled, they work perfectly, and get full bandwidth available from the same peers. Upload is not affected. Two different torrent clients do not have any problems with TCP connection speed with the same torrents on the same system. Multiple torrents with big and small swarms were tested. Total speed is proportional to the number of connected peers, but each one never goes over that small limit.

Most settings are at their defaults. Priorities, bandwidth quotas, piece creation limiter were left untouched. A portable Tixati version was cleanly installed and tested, and behaved the same. No VPN is used, and there is abundance of available network bandwidth.

It is possible that some external force (some Linux TCP option processing? some broken ISP traffic shaper?) is messing with TCP packet headers, and causes that bandwidth cap just for Tixati, but there might also be some internal bug that causes request queue to get stuck at small values, and ignore the rest.

More tests incoming.

User tried to run Tixati on Mint, Fedora, and EndeavourOS, with multiple stable and fresh kernels provided (6.12, 6.17, 6.18, 6.19, and so on). bbr and cubic settings for net.ipv4.tcp_congestion_control were tested. Peers still get stuck at certain speed in Tixati only.

I've tried to reproduce it myself on regular Debian stable on a bootable flash, and results are suspicious.

https://files.catbox.moe/2c127g.png

This is not me setting the bandwidth limit. This is a regular torrent with 1, 2 or 3 TCP peers (from various countries) allowed to connect. They almost always stick to 21/21, 24/24 or lesser values divisible by 3 in queue, and send data at 140 KB/s or less. The only peers who seem to be able to jump to high speeds are the ones from my country, presumably with low ping and fast acknowledgements. I guess there might be some problem with flushing the data down the receive buffers.

Windows build running under Wine does not have the problem.