Hi there,
I’ve got a pool with sync disabled, and an NFS client connected via 10Gb Ethernet. When I mount and do a dd test, it always maxes out at 105MB/s (840Mbps). When I run the same test locally on TrueNas, I get 3.0GB/s (which I expect), but I would still be hoping to see 1200-1000MB/s via NFSv4 mount.
Any advice or tips?
Are you running CORE or SCALE?
What mount options are you using on the client, and what is the client?
Did you do iperf test? 105MB/s is highly suspicious of link negotiating at 1Gb not 10Gb.
Scale Dragonfish.
Client is RHEL 9.3 with default options
192.168.41.1:/my-nfs-mount /my-nfs-mount nfs4 rw,defaults 0 0
iperf3 confirms 10Gb/s.
None from my end since we cant reproduce such results.
Have you tried SMB to pinpoint the problem as a NFS issue?
Have you tried setting the nconnect options on the client mount side?
You may want to experiment with tuning on that side first, it can have a significant impact on NFS performance. Note, nconnect should only be used with TrueNAS SCALE NFS servers. It may not be data safe for CORE.
That worked! However, I wasn’t successful with simply adding nconnect=16 into the options section of /etc/fstab. But, I manually mounted with mount -f nfs ... and the mount option appeared.
One more interesting observation… nconnect=16 doesn’t appear to work with NFSv4.2, but only v4.1. So, in my /etc/fstab, I had to tie it to that version by adding vers=4.1. Then, mounting the path with fstab does show that the nconnect option was included in the mount. For reference I’m using kernel 5.14.0-362.24.1.el9_3.0.1.x86_64
