Re: Low capped TCP network speeds over the Internet

[z411 <z411%omaera.org@localhost>]

Good news again!

I managed to increase client->server scp speeds to the expected 8 MB/s
by specifically increasing recvspace:

net.inet.tcp.recvspace=4194304

From what I understand, the TCP window should've been growing
automatically, and since my recvbuf_max is already 16MB, I didn't think
recvspace would be the bottleneck. Apparently, that's not the case when
SSH receives data?

Previously, both recvspace and sendspace were 262144, with both having a
max of 16777216. Other traffic managed to scale and use the maximum size
fully, including SSH when sending. But for some reason, SSH requires a
larger initial recvspace when receiving. It's the only exception so far.

Either way, here are my current tunables:

kern.sbmax=16777216
net.inet.tcp.recvspace=4194304
net.inet.tcp.sendspace=4194304
net.inet.tcp.recvbuf_auto=1
net.inet.tcp.sendbuf_auto=1
net.inet.tcp.recvbuf_max=16777216
net.inet.tcp.sendbuf_max=16777216
net.inet.tcp.recvbuf_inc=262144
net.inet.tcp.sendbuf_inc=262144
net.inet.tcp.congctl.selected=cubic

SSH, HTTP and iperf3 are now performing very well in both directions.

I think I have no more complaints. I'm definitely ready to put NetBSD
into production and I'm pretty happy, because I fell in love with the
system the moment I first installed it.

I want to write a follow-up tomorrow in my blog so people don't wrongly
assume there's something inherently wrong with NetBSD's networking stack.

Thank everyone who helped me, I'm definitely a NetBSD fan now.
Best regards,
- z411