Establish what you can control

A KVM guest usually allows guest-kernel administration, while a container host may restrict algorithms and sysctl settings. Inspect the environment before replacing a kernel or disabling network protections in the name of speed.

uname -r
sysctl net.ipv4.tcp_congestion_control
sysctl net.ipv4.tcp_available_congestion_control
sysctl net.core.default_qdisc
ip route show default
tc qdisc show

Confirm that the workload uses TCP. BBR does not directly control UDP/QUIC congestion handling, remove upstream rate limits, fix CPU saturation or expand transfer quotas. If BBR is unavailable, consult the distribution and hosting documentation instead of assuming that every VPS can load it.

Record defaults, then try a temporary change

Save the original algorithm and queue discipline in your operations notes. With kernel support and a recovery console available, a limited experiment can use:

sudo modprobe tcp_bbr
sudo sysctl -w net.ipv4.tcp_congestion_control=bbr
sudo sysctl -w net.core.default_qdisc=fq
sysctl net.ipv4.tcp_congestion_control

Stop if the module cannot load. The default congestion algorithm mainly affects new connections; existing transfers are not a clean before-and-after experiment. Changing default_qdisc also does not guarantee replacement of an already configured interface queue. Check tc qdisc show and avoid blindly deleting queue rules or restarting networking through your only remote session.

Compare the same workload

Use a receiver you administer or are authorized to test. Keep the target, period, transfer size and connection count comparable. Record CPU load, retransmissions, sustained throughput and interactive latency, not just a peak number. Single-stream and multi-stream tests differ, and the network changes over time.

If results worsen, restore the recorded algorithm and default qdisc and establish new test connections. Do not assume that the original values must be cubic and fq_codel. Only persist verified changes in a separate /etc/sysctl.d/90-vpsspy-bbr.conf after testing; rollback removes that file and restores the recorded settings.

Avoid universal tuning bundles

A large copied set of buffer changes can increase memory use, queueing latency or maintenance difficulty. Change a setting only when a measurement identifies the bottleneck, and keep the reason, test and rollback with it.

If the issue is routing or peak-hour congestion, compare locations and routes. If resources are insufficient, check CPU and memory with the finder. Kernel tuning does not replace selecting suitable compute and connectivity.