On Thu, 14 Mar 2024 at 12:22, Melih Mutlu <m.melihmu...@gmail.com> wrote: > I did some experiments with this patch, after previous discussions
One thing I noticed is that the buffer sizes don't seem to matter much in your experiments, even though Andres his expectation was that 1400 would be better. I think I know the reason for that: afaict from your test.sh script you connect psql over localhost or maybe even unix socket to postgres. Neither of those would not have an MTU of 1500. You'd probably want to do those tests over an actual network or at least change the MTU of the loopback interface. e.g. my "lo" interface mtu is 65536 by default: ❯ ip a 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo valid_lft forever preferred_lft forever