I am migrating about 40TB+ from a Synology NAS to Unraid. I have the mounted the Synology on Unraid with unassigned devices.
However the transfer speed averages around 30MB/s no matter what I tried. I think it might be a problem with SMB or networking.

All devices are connected to a single managed switch.

 

Things that I have tried:

1. I have tried both with cache enabled and cache disabled on the target user share.

2. I have also tried to transfer to a cache-only folder (appdata folder) but speed also remains the same (ave 30MB/s).

3. I ran iperf3 between Unraid, Synology NAS, Router (pfsense), as well as my computer. All tests achieve 950+ MB/s. So it's not a network problem.

4. Tested array with DiskSpeed container. Slowest drive average 138 MB/s and fastest drive average 286 MB/s with a simultaneous max throughput of 1.52 GB/s. Therefore it is not a drive or the controller problem.

5. Tried copying with Krusader, Double Commander, as well as Dynamix File Manager. No difference in speed. 
6. In Tips and Tweaks, I have tried (one at a time) disbling NIC Flow Control and NIC Offload. I have also increased Rx and Tx buffer to 1024. Still no changes to transfer speed. 
7. Previously I had this Samba extra configuration:

#server multi channel support = no
server min protocol = SMB3_02
client ipc min protocol = SMB3_11
client NTLMv2 auth = yes
restrict anonymous = 2
null passwords = no
raw NTLMv2 auth = no

#vfs_fruit compatibility for Apple SMB
[global]
vfs objects = catia fruit streams_xattr
fruit:nfs_aces = no
fruit:zero_file_id = yes
fruit:encoding = native
spotlight backend = tracker

fruit:metadata = stream
fruit:posix_rename = yes
readdir_attr:aapl_max_access = no
readdir_attr:aapl_finder_info = no
readdir_attr:aapl_rsize = no

but now I have also removed all of it and Samba extra configuration is empty.
8. I have tried toggling Tunable (enable Direct IO) and reconstruct write.
9. I have also tried removing parity drive and writing directly to the unprotected array. It made no difference to the transfer speed. So I have now rebuilt parity and chose to do the migration with parity protection.

10. I also tried mounting the Unraid share in Synology and doing the transfer on Synology. But the speed was even worse at around 20MB/s.

11. Tried mounting the Synology share using NFS instead of SMB but it made no difference. 
12. Doing a transfer speed test from my Mac to Unraid using Blackmagic Disk Speed Test always give random results. Sometimes it achieve full 100MB/s but most other times it is way less around 10-20 MB/s. The same test to Synology share always gets 100MB/s.
 

The only thing I am hesitant to try is modifying docker network setting as it would require changing all of my container network settings to match. Currently I have macvlan + host access to custom network enabled.  

 

I am out of ideas of what to do. Please give me some suggestions on what else to try as 40TB+ transfer at 30MB/s would take a couple weeks instead of days. 

tower-diagnostics-20240906-1019.zip

Edited September 6, 20241 yr by Sak

Disk1 is SMR, try transferring to disk2, without parity or with turbo write enabled.

29 minutes ago, JorgeB said:

Disk1 is SMR, try transferring to disk2, without parity or with turbo write enabled.

I use disk 1 for seeding torrents only. It is excluded from the share that I am moving files into. 
 

I have also tried removing parity and turbo write as well. Made no difference.
 

Have you actually set the shares to have Exclusive Access?

 

 

I'm trying to think back to my Synology migration to Unraid...I think I went for rsync via ssh rather than using a Samba or NFS share mounted in either direction.  Anyway I noticed a Samba performance increase post migration after setting the above.

 

https://unraid.net/blog/6-12-0-stable

5 minutes ago, KillerK said:

Have you actually set the shares to have Exclusive Access?

 

 

I'm trying to think back to my Synology migration to Unraid...I think I went for rsync via ssh rather than using a Samba or NFS share mounted in either direction.  Anyway I noticed a Samba performance increase post migration after setting the above.

 

https://unraid.net/blog/6-12-0-stable

 

I'm transferring to a "Media" share spread over 4 drives (disk2 to disk5). Since it is spread over 4 drives it does not have exclusive access. The share has cache enabled but I am transferring to /mnt/user0 to bypass cache. 

However, when I tested copying a file to "appdata" share that is cache-only and have exclusive access, the transfer speed still hovers around 30MB/s which is puzzling to me as it is writing directly to the ssd cache pool. 

I did not know you could rsync over ssh, thanks for the suggestion, I will have to try that. Since iperf between the Synology and Unraid was full gigabit, I think I will see a speed improvement with rysnc.

29 minutes ago, Sak said:

 

I'm transferring to a "Media" share spread over 4 drives (disk2 to disk5). Since it is spread over 4 drives it does not have exclusive access. The share has cache enabled but I am transferring to /mnt/user0 to bypass cache. 

FWIW I've 4 disks in a zpool with a samba share built ontop and this has exclusive access no issues.

24 minutes ago, KillerK said:

FWIW I've 4 disks in a zpool with a samba share built ontop and this has exclusive access no issues.

 

I just tried disabling cache on the share again and it does not have exclusive access. I think it's because you use zpool so you have a single unified pool similar to how Synology handles the volumes. I just use the regular xfs as I want to mix and match the disk as well as to spin down the disk when not in use. 

36 minutes ago, KillerK said:

FWIW I've 4 disks in a zpool with a samba share built ontop and this has exclusive access no issues.

This is because it is a pool with a single file system.   When in the array each disk is a separate file system.

I just tried rsync, I get around 20MB/s

1 hour ago, Sak said:

I just tried rsync, I get around 20MB/s

ouch, can you test a zpool to get an exclusive access share to prove/disprove its your disk and share layout?

Edited September 6, 20241 yr by KillerK

You mentioned running  iperf, please post the actual results of a single stream test in both directions.

2 hours ago, JorgeB said:

You mentioned running  iperf, please post the actual results of a single stream test in both directions.

 

iperf test to Synology NAS

 

copy direct to /mnt/cache of "ephemeral" – a cache-only share 

 

copy direct to /mnt/cache of "Media" – the target folder

 

copy to /mnt/user 

 

copy to /mnt/user0 (I think this is FUSE but bypass cache) which is my intended way of copying all the files.

 

Same as above but using double commander. Better speed here.

 

Lastly a copy using double commander directly to /mnt/cache/Media 

 

Drives and Controller benchmark

 

Could the issue be the transmit end on your Syno? Any kind of DPI on the egress?

3 minutes ago, KillerK said:

Could the issue be the transmit end on your Syno? Any kind of DPI on the egress?

I do have DPI on the router, but since all the devices are connected on a single switch, LAN traffic does not pass through the pfsense router. 

 

Running a speed test from my Mac to the Synology also shows no problem on the Synology side. 

Try this and post the results, in the Unraid CLI type:

 

pv /mnt/remotes/path/to/large/file > /dev/null

 

29 minutes ago, JorgeB said:

Try this and post the results, in the Unraid CLI type:

 

pv /mnt/remotes/path/to/large/file > /dev/null

 

To me that confirms the source is the problem, do you have a PC that you could use to make an SMB transfer to Unraid?

10 hours ago, JorgeB said:

To me that confirms the source is the problem, do you have a PC that you could use to make an SMB transfer to Unraid?

 

Tested with a spare windows laptop. Seems like the issue is writing into Unraid. No problem with reading or with the Synology NAS.

 

From Unraid to Desktop:


From Desktop to Unraid. Copying to "Media" share – a share with Cache as primary and Array as secondary. 


From Synology to Desktop


From Desktop to Synology

Edited September 7, 20241 yr by Sak

4 hours ago, Sak said:

Seems like the issue is writing into Unraid.

The test with pv confirmed the problem was not writing to Unraid, since it was sending the data to /dev/null and it was still slow, but if it's also slow with the laptop, it suggests a network issue, despite the good iperf results, do you have a different NIC you could test with? I also would try a direct connection from the laptop to Unraid, to rule out cable/switch issues.

1 hour ago, JorgeB said:

The test with pv confirmed the problem was not writing to Unraid, since it was sending the data to /dev/null and it was still slow, but if it's also slow with the laptop, it suggests a network issue, despite the good iperf results, do you have a different NIC you could test with? I also would try a direct connection from the laptop to Unraid, to rule out cable/switch issues.

 

I think you're right that it's a network issue. Looking at the Errors Info, I'm seeing a lot of drops and overruns for receive counters.

 

It seems like flow control was enabled for Rx but not for Tx.

 

In tips and tweaks, Flow Control is enabled by default.

 

Do you recommend to enable or disable flow control for both Rx and Tx?

Edited September 7, 20241 yr by Sak

Comparison for you...

 

~# ethtool -a eth0
Pause parameters for eth0:
Autonegotiate:  on
RX:             on
TX:             on

12 hours ago, KillerK said:

Comparison for you...

 

~# ethtool -a eth0
Pause parameters for eth0:
Autonegotiate:  on
RX:             on
TX:             on

 

Thanks for the help so far. After switching out cables and testing all network equipments, it is definitely an issue with the Ethernet controller: Intel Corporation Ethernet Controller I226-V (rev 06) on my motherboard. 
 

Another user had a problem with this NIC as well. 

 

Officially VLANs are not supported on this ethernet controller. However my setup had a vlan enabled in network settings. I thought that might be the cause of the issue but even after I deleted all the VLANs, the issue still persists. 

Another odd thing I noticed is that flow control for Tx cannot be turned on. I have tried enable/disable through both "Tips and Tweaks" as well as using the ethtool -A commands. Flow control for Autonegotiate and Rx can be turned on and off, however Tx always remains off. 
In addition, I noticed that when I have turned off flow control for this NIC, sometimes Rx would be on by itself:
 

Autonegotiate:	off
Rx:		on
Tx:		off

 

 

I also did a direct connection from my Mac to Unraid and did multiple iperf test. The Unraid to Mac test always get full speed but the Mac to Unraid was not consistent. About half the time it gets full speed, and the rest of the time was unstable with around 300Mbps. 

 

 

For now I am using a USB ethernet adapter plugging directly into the second port of the Synology. I am able to get full transfer speed with this method. 

 

 

is it possible to install our own driver for the I226V? https://www.intel.com/content/www/us/en/download/15084/intel-ethernet-adapter-complete-driver-pack.html

The USB Ethernet adapter is not an ideal permanent solution. 

3 hours ago, Sak said:

is it possible to install our own driver for the I226V?

Not that I know of.

2 hours ago, JorgeB said:

Not that I know of.

 

A quick google search seems to reveal a lot of stability issues with i225 and i226. Hopefully it is fixed soon through the IGC driver updates.