Poor performance when using NFS share w/ Docker Container

[tronbo]

Hello everyone,

I have two servers, both of which Proxmox is the hypervisor. On one machine, I have Truenas virtualized in a VM. On another, I have a Ubuntu VM which hosts a few docker containers. One of the containers is sabnzbd, and I am using an NFS v4.1 share to store usenet downloads on the truenas VM.


Specs (VM):
HDD: 8x 8TB Seagate Ironwolf (oculink PCIE hardware passthrough)
CPU: 8 cores, AMD Ryzen 7 3700x
RAM: 28GB
NIC: x1 10GLAN Intel X550-AT2 , internal network is capped at 1Gbps :-(

Pool:
8x 8TB 7200RPM disks
raidz2
Size (total): 40,08 tb

Lastly, here is how I am establishing the NFS connection in the docker-compose.yml file:

Code:

volumes:
    media:
        driver: local
        driver_opts:
            type: nfs
            o: nfsvers=4,addr=192.168.1.84,rw,async
            device: ":/mnt/chungus/media"

Initial Tests:


1. From sabdnzbd container (NFS share v4)

It really struggles to go above 40-40MB/s and dips down to 18-20MB/s. It ramps back up, though. I should be getting 90MB/s +


2. From mac (SMB share)

Here, you can see that the only bottleneck is my line speed. I'd love to get this performance out of my NFS share.


I'm not sure what other tests I should be making to find where the bottle neck could be coming from? Any advice on where I should start would be much appreciated!

Cheers!

 

[1nv4lidus3r]

Was this ever resolved. I have a very similar issue

 

[Jessep]

Is NFS on Proxmox/VM/TrueNAS synchronous or asynchronous transfers?

 

[Bern]

My guess would be that you only have 28GB of RAM for a 64TB pool.

 

[HoneyBadger]

Is NFS on Proxmox/VM/TrueNAS synchronous or asynchronous transfers?

IIRC the TrueNAS NFS server defaults to synchronous write behavior (which is the "correct" or "expected default") and this would slow things down especially with a RAIDZ backing pool.

 

[NugentS]

try setting sync=disabled and see if that speeds things up by running the same tests again.
WARNING: This setting is considered non data-safe in that if you have a power event / kernel panic you may lose data in flight. However for testing it would likley point at the issue.