Hello everyone,

I’m new to the UNRAID world, and this is my first experience with the system, which I find amazing. I immediately purchased the Unleashed version. However, I’m experiencing major performance issues and can’t resolve them, so I’m asking for assistance from this very active community!

 

Problem Summary

With some processes/services, I can transfer files over LAN and between server disks at decent/expected speeds (around 10 Gb/s on LAN and 1.80 GiB/s between NVMe drives).

With other services, speeds drop drastically, to about 150 MB/s or even 5–10 MB/s inside certain Docker containers.

 

Server Configuration:

 

Hardware:

Model: HPE Proliant DL380 G9 (24x 2.5" SF)

CPUs: 2x Xeon E5-2667 V3, 8 cores each @ 3.20 GHz

RAM: 384GB DDR4 (24 x 16GB)

 

Drives:

19 x 1.8TB 10K SAS (12Gb/s)

2 x 480GB SATA SSD

2 x 2TB NVMe (Gen4)

1 x 4TB NVMe (Gen4)

 

Storage Configuration:

Array: 17 SAS drives + 2 parity drives (XFS)

Cache Array: 2 SSDs in RAID1 (Btrfs)

Pool 1 (nvme-raid): 2 NVMe drives (2TB each) in RAID1 (Btrfs)

Pool 2 (nvme-singolo): 1 NVMe drive (4TB) (XFS)

 

Additional Info

Unraid Version: v7 stable

Network Card: 10 Gb fiber

Internet Connection (ISP): 10 Gb download, 2.5 Gb upload

Various Options Tested (enabled/disabled):

Permit exclusive shares: YES

Tunable (md_write_method): reconstruct write

Samba extra configuration:

[global]
server multi channel support = yes
use sendfile = yes
min receivefile size = 16384
aio read size = 1
aio write size = 1
socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=131072 SO_SNDBUF=131072
max protocol = SMB3
read raw = yes
write raw = yes

 

Test Summary

1. Ookla Speed Test from Unraid CLI (OK)

Server: TIM SpA - Milan (id: 3667)
ISP: TIM Business
Download:  8127.59 Mbps (data used: 9.8 GB)
Upload:    2086.88 Mbps (data used: 2.1 GB)
Result URL: https://www.speedtest.net/result/c/7edff5bc-9268-4d48-9fb5-9bc4ef62c98a

 

 

2. Ookla Speed Test from Docker (jDownloader) (OK)

• Same performance as the CLI.
   

3. File Downloads from jDownloader (Docker) (KO)

• Starts fast, then drops to ~10 MB/s.

 

4. Copy from Windows 11 PC (10 Gb NIC, NVMe) to an Unraid NVMe Share (KO)

• Transfer doesn’t exceed 160 MB/s.
• On other servers, I can easily reach 900 MB/s or more.

 

5. Local VM on NVMe, SMB speed depends on Virtio driver

• Virtio-net: max ~150 MB/s
• Virtio: ~990 MB/s (full bandwidth)

 

6. Transfers between NVMe disks using rsync (KO)

rsync -ah --progress --info=stats1 \
/mnt/nvme-singolo/Downloads/TESTFILE.mkv /mnt/nvme-raid/
21.98G 100%  367.99MB/s in 56s (~382 MB/s effective)

 

7. Transfers between NVMe disks using pv (OK)

pv /mnt/nvme-singolo/Downloads/TESTFILE.mkv > /mnt/cache/TESTFILE.mkv
20.5GiB 0:00:16 [1.27GiB/s] [============================>] 100%

 

8. iPerf Test between Windows 11 PC and Unraid (OK)

[SUM]   0.00-10.00 sec  11.6 GBytes  9.92 Gbits/sec  sender
[SUM]   0.00-10.01 sec  11.5 GBytes  9.91 Gbits/sec  receiver

 

 

In Summary

 

Internet speed test: OK

• Transfers between NVMe/SSD using Unraid GUI or rsync: slow
• Transfers between NVMe/SSD using pv: fast
• SMB transfers from local VM to external server: speed varies (depends on Virtio driver)
• jDownloader (Docker): ~10 MB/s (very slow)
• SMB transfers to/from Unraid: ~160 MB/s (underwhelming for a 10 Gb link)
   

Thanks in advance to anyone who can help solve this mystery!

unbox-diagnostics-20250119-1046.zip

Edited January 19, 20251 yr by maxita
attached diagnostics as requested by moderator

Attach Diagnostics to your NEXT post in this thread.

 

For transfers involving writes to the parity array, see here:

https://docs.unraid.net/unraid-os/manual/storage-management/#array-write-modes

4 minutes ago, trurl said:

Attach Diagnostics to your NEXT post in this thread.

 

For transfers involving writes to the parity array, see here:

https://docs.unraid.net/unraid-os/manual/storage-management/#array-write-modes

@trurlsorry for forgetting, I attached the diagnostics to the post as requested.

I would like to point out that the problem is not only with the array but also with the nvme disk pools where there is no active array.

In your diagnostics, logs/syslog.1.txt and logs/syslog.2.txt

Jan 19 03:57:32 UNBox sshd[4063873]: drop connection #0 from [185.147.124.54]:40616 on [192.168.1.243]:22 penalty: exceeded LoginGraceTime
Jan 19 03:59:48 UNBox sshd[4063873]: drop connection #3 from [92.255.85.253]:46456 on [192.168.1.243]:22 penalty: failed authentication

And lots of similar.

 

https://www.abuseipdb.com/check/92.255.85.253 Russia

https://www.abuseipdb.com/check/185.147.124.54 Russia

4 hours ago, trurl said:

In your diagnostics, logs/syslog.1.txt and logs/syslog.2.txt

Jan 19 03:57:32 UNBox sshd[4063873]: drop connection #0 from [185.147.124.54]:40616 on [192.168.1.243]:22 penalty: exceeded LoginGraceTime
Jan 19 03:59:48 UNBox sshd[4063873]: drop connection #3 from [92.255.85.253]:46456 on [192.168.1.243]:22 penalty: failed authentication

And lots of similar.

 

https://www.abuseipdb.com/check/92.255.85.253 Russia

https://www.abuseipdb.com/check/185.147.124.54 Russia

Yes thanks @trurl I have enabled 2 days ago ssh for connect from home without vpn. Now I have changed rules on firewall for accept just my home ip to the ssh port.

Hey everyone,

 

Quick update on my previous SMB transfer speed issues. I discovered that disabling the Tailscale plugin entirely resolved most of my SMB bottlenecks. After removing Tailscale and all related virtual interfaces, I’m now seeing full 10 Gbit throughput when copying via SMB—whether I use user shares or disk shares.

 

It seems that Tailscale and its virtual network interfaces were introducing a bottleneck for the SMB protocol, so if you’re having similar speed problems and happen to be running Tailscale, try disabling it to see if it helps!
 

However, I still have one lingering problem: the jDownloader Docker containers are stuck around 10 Mbit/s download speeds, which is significantly slower than expected. I tested by installing the Speedtest CLI inside the container:

• The first Speedtest sometimes reports a relatively good download rate.

• But subsequent Speedtests (and real downloads via jDownloader) plummet to very low throughput—as if there’s some sort of QoS or bandwidth cap kicking in.

 

On my firewall side, I have no traffic shaping or throttling configured. The same version of jDownloader running on Windows in the same network saturates my connection without any issues. So it’s definitely isolated to the container environment on Unraid.

 

Any ideas on what could be throttling the Docker container’s traffic? If anyone has come across a similar jDownloader Docker speed cap or any Docker networking QoS problem, please let me know. I’ve tried all the usual suspects (disabling Tailscale, checking container network modes, etc.) without success.

 

Thanks in advance for any tips! @trurl any ideas?
 

UPDATE

 

seems a problem with only jDownloader containers as you can see here:

 

UPDATE: i have fixed the problem of jDownloader forcing the MTU of docker to 1500 also if my network works well at 9000 (10Gbit network)

But…

After several tests I came to the conclusion that UNRAID 7 has some bug that limits the speed of writing and reading the disks. Not because of FUSE because I tried to bypass it but without improvements. I am working on nvme disks not on arrays.
When I use commands like DD the disks go at high speeds even using real files. When I use the unraid file manager I go at least half the speed if not less.

When I transfer via SMB on a 10 Gbit network after a thousand tweaks I managed to transfer at 500/600 MB/s. The strange thing is that if I change the MTU of UNRAID and the windows client to 9000 for a few hours I transfer at 1.2 GB/s and then align again at 500/600 MB/s.

Something inside unraid caps the speeds.

17 hours ago, maxita said:

UPDATE: i have fixed the problem of jDownloader forcing the MTU of docker to 1500 also if my network works well at 9000 (10Gbit network)

But…

After several tests I came to the conclusion that UNRAID 7 has some bug that limits the speed of writing and reading the disks. Not because of FUSE because I tried to bypass it but without improvements. I am working on nvme disks not on arrays.
When I use commands like DD the disks go at high speeds even using real files. When I use the unraid file manager I go at least half the speed if not less.

When I transfer via SMB on a 10 Gbit network after a thousand tweaks I managed to transfer at 500/600 MB/s. The strange thing is that if I change the MTU of UNRAID and the windows client to 9000 for a few hours I transfer at 1.2 GB/s and then align again at 500/600 MB/s.

Something inside unraid caps the speeds.

I am having the same issue with NVME cache on Unraid. My Sabnzbd reports write speed of 500MB. It should be well in excess of 2K-3K/sec
Ignore the temp download as that is a ram disk. 
My system is pretty beefy with 192GB ram and my internet is 6Gbps symetric but i'm being limited by my NVME speeds atm. The cache drive is dedicated share.