IPFS uses QUIC transfers. These connections can be limited by the size of the UDP receive buffer. This buffer holds packets that have been received by the kernel, but not yet read by the application (quic-go is what IPFS uses in this case). Once this buffer fills up, the kernel will drop any new incoming packet.
ensure you have configured
sysctl -w net.core.rmem_max=2500000
default is 208 kiB
you want: 2048 kiB
see https://github.com/lucas-clemente/quic-go/wiki/UDP-Receive-Buffer-Size
IPFS uses QUIC transfers. These connections can be limited by the size of the UDP receive buffer. This buffer holds packets that have been received by the kernel, but not yet read by the application (quic-go is what IPFS uses in this case). Once this buffer fills up, the kernel will drop any new incoming packet.
ensure you have configured
default is 208 kiB
you want: 2048 kiB
see https://github.com/lucas-clemente/quic-go/wiki/UDP-Receive-Buffer-Size