Re: Poor TCP performance as latency increases

[Havard Eidnes <he%NetBSD.org@localhost>]

> It seems that there is a performance issue in the TCP stack,
> and I'm hoping someone else can reproduce my findings as I'm
> not an expert in this area.
>
> Simple test: Download a file from a web server. As latency
> between the client and server increases, there is a drastic
> reduction in throughput. I don't see this with other (Linux,
> macOS) TCP stacks.

Traditionally, NetBSD has been fairly conservative in the
parameter settings for the TCP stack, meaning "not allocating too
much memory".

If you find that your performance scales inversely with RTT, that
is probably the issue: your TCP sessions will then be "window
size limited".

You could test with these settings from /etc/sysctl.conf and see
if they make any discernible difference:

# Tuning of TCP, slightly conservative:
kern.sbmax=1048576
net.inet6.tcp6.sendbuf_max=1048576
net.inet6.tcp6.recvbuf_max=1048576
net.inet.tcp.sendbuf_max=1048576
net.inet.tcp.recvbuf_max=1048576
net.inet.tcp.congctl.selected=cubic

# And also bump the default receive window:
net.inet.tcp.recvspace=262144

I also agree with Greg Troxel that to fully diagnose issues such
as this, you ideally need pcap packet traces, and minimally at
the sending side and use tcptrace + xplot to inspect the result
to see if there are retransmissions (which is indicative of
packet loss, which is bad for performance but is also the way the
Internet works to provide feedback to the TCP flow control when
there is congestion), or whether you are window-size-limited or
whether there is some other issue.

Regards,

- Håvard