First I want to apologize for how much of a beginner I am😅.

I have setup an array of 6 disks, 1 of which is a parity drive. When copying 2 single large files from the array to my windows machine I notice the speed is halved even though the files are copying from 2 different drives in the array. I have a 10Gbe network card on both ends. So I will be getting about just over 200MB/s and then as soon as i start copying another file I notice each file copies at just under 100MB/s. I would understand this if they was copying from the same drive but I thought this wouldnt be the case as they are copying from 2 different drives. Any help with this would be much appreciated.

Unraid Spec:

Intel Core Ultra 5 225

32GB DDR5 Ram

ASRock Z890I Nova WiFi Motherboard

6x Toshiba 12TB 7200rpm Drives

2x 500GB Samsung 970 NVME Cache (btrfs)

ASUS XG-C100C 10G Network Adapter PCI-E x4

Edited August 28, 2025Aug 28 by NSDiesel
spelling

There's typically a little slowdown when reading from multiple disks, but not as much. Is it any different if you enable disk shares and copy from two disks using those?

Thanks for the reply. I will give this a try and see if anything changes. Im also thinking maybe it is my M.2 NVME to 6x SATA adapter that is a bottleneck.

56 minutes ago, NSDiesel said:

Thanks for the reply. I will give this a try and see if anything changes. Im also thinking maybe it is my M.2 NVME to 6x SATA adapter that is a bottleneck.

So just tried using disk shares and got the same result.

I have added two screenshots to show the difference in speed and as I said both files are on different drives in the array.

5 hours ago, NSDiesel said:

my M.2 NVME to 6x SATA adapter that is a bottleneck.

It should not be if operating normally, but you can run the diskspeed container to test the total controller bandwidth.

16 hours ago, JorgeB said:

It should not be if operating normally, but you can run the diskspeed container to test the total controller bandwidth.

So I ran a bandwidth test using diskspeed on my controller. The rsult was "Current & Maximum Link Speed: 8GT/s (PCIe 3) width x2 (1.97 GB/s max throughput)"

Would 6x Sata 7200rpm HDD's bottleneck this?

It shouldn't, but you can run an actual controller bandwidth test with that app, where all disks are read simultaneously.

Based on the results below it does not look like the controller is the problem. Not sure what else to try to see what the isue might be

Yep, that looks fine. Please post the diagnostics so that I can look at your config to see if I can reproduce it.

Diagnostic file attached.

Thanks for looking into this

nsd-unraid-diagnostics-20250830-1933.zip

I've retested this, and basically I'm seeing the same as you: 1.iso comes from disk1, 2.iso comes from disk2:

I had reported this to LT more than a year ago, but at the time, it wasn't so noticeable in my testing with fewer devices, especially with XFS.

Tests at the time were done with SSDs and don't remember exactly how I tested, but it wasn't over SMB, so that can also play a part.

I brought this up again to LT, since it's much more noticeable in my current tests, so hopefully something can be done in the future to help with this.

Thanks for making LT aware. I guess we will just have to see what happens going forward with updates.

I might give it a try with NFS shares ans see if that makes a difference l.

So after discussing this with LT and doing some more testing, it does look as if it may be more of an SMB issue, I see the same speed halving if I copy from two independent pools, and those don't use the md driver.

Please retest locally with pv to /dev/null like I did, to see if you get the same results.

Use a large file as source and run both copies from two different disks at the same time, I've also tested from two different pools at the same time, and it was the same:

Note that if you do multiple runs, always stop and restart the array before each run, or you will likely get bad results due to RAM caching.

P.S. Different disks were used this time, hence the different speeds.

1 hour ago, JorgeB said:

So after discussing this with LT and doing some more testing, it does look as if it may be more of an SMB issue, I see the same speed halving if I copy from two independent pools, and those don't use the md driver.

Please retest locally with pv to /dev/null like I did, to see if you get the same results.

Use a large file as source and run both copies from two different disks at the same time, I've also tested from two different pools at the same time, and it was the same:

Note that if you do multiple runs, always stop and restart the array before each run, or you will likely get bad results due to RAM caching.

P.S. Different disks were used this time, hence the different speeds.

Definetly much faster for me when running your test scenario.

Thanks for looking into this for me. I will use the web terminal whenever i need the extra speed as this seem much faster.

Some more findings: this is reading from two SSDs over SMB. The first copy starts at around 390MB/s, and it gets halved when starting the 2nd one:

After adding these to SMB extras in Unraid (Settings - SMB - SMB Extras)

aio read size = 1

aio write size = 1

So this appears to be the resolve the issue; please try.

1 hour ago, JorgeB said:

Some more findings: this is reading from two SSDs over SMB. The first copy starts at around 390MB/s, and it gets halved when starting the 2nd one:

After adding these to SMB extras in Unraid (Settings - SMB - SMB Extras)

aio read size = 1

aio write size = 1

So this appears to be the resolve the issue; please try.

So I tried the same but for me it did not work. It seems to have slowed down copying a single file to less then 100MB/s and then just under half that when copying 2 files from different disks.

I then removed the text from SMB extras and then it speeded up again but still get half the speed when copying 2 files.

That's unexpected, but testing with HDDs instead, I also see the same: speeds are slower than without AIO, but curiously, if I pause/unpause the transfers in Windows Explorer, they do then go faster:

Very strange issue.

2 hours ago, JorgeB said:

That's unexpected, but testing with HDDs instead, I also see the same: speeds are slower than without AIO, but curiously, if I pause/unpause the transfers in Windows Explorer, they do then go faster:

Very strange issue.

Same for me. I paused and then unpaused and the transfer went to normal speed. That is a very strange one!😅

6 minutes ago, NSDiesel said:

That is a very strange one!

I've found since that the issue is also resolved with AIO and HDDs if they are used in two different pools instead of the array:

It appears to me that there's some interaction between the md driver (for array HDD devices) and the Samba AIO option:

To recap:

• with AIO disabled, everything works fine for a single transfer, but if users try to do multiple transfers over SMB from different devices, array or pools, read speed will be approximately halved.

• with AIO enabled, HDDs (with pools only) and SSDs (with pools or array) no longer see this speed halving issue for multiple transfers over SMB.

• with AIO enabled, HDDs in the array don't perform well over SMB, and will typically be slower than with AIO disabled, even for single transfers (pausing and unpausing Windows Explorer may help, but it's not really a solution).

I've forwarded this info to LT, maybe they can do something about it in the future.

If the issues with array HDDs and Samba AIO could be resolved, they could enable it by default.

On 9/2/2025 at 10:50 AM, JorgeB said:

I've found since that the issue is also resolved with AIO and HDDs if they are used in two different pools instead of the array:

It appears to me that there's some interaction between the md driver (for array HDD devices) and the Samba AIO option:

To recap:

• with AIO disabled, everything works fine for a single transfer, but if users try to do multiple transfers over SMB from different devices, array or pools, read speed will be approximately halved.

• with AIO enabled, HDDs (with pools only) and SSDs (with pools or array) no longer see this speed halving issue for multiple transfers over SMB.

• with AIO enabled, HDDs in the array don't perform well over SMB, and will typically be slower than with AIO disabled, even for single transfers (pausing and unpausing Windows Explorer may help, but it's not really a solution).

I've forwarded this info to LT, maybe they can do something about it in the future.

If the issues with array HDDs and Samba AIO could be resolved, they could enable it by default.

Thanks for all your help with this👍

On 9/2/2025 at 10:50 AM, JorgeB said:

I've found since that the issue is also resolved with AIO and HDDs if they are used in two different pools instead of the array:

It appears to me that there's some interaction between the md driver (for array HDD devices) and the Samba AIO option:

To recap:

• with AIO disabled, everything works fine for a single transfer, but if users try to do multiple transfers over SMB from different devices, array or pools, read speed will be approximately halved.

• with AIO enabled, HDDs (with pools only) and SSDs (with pools or array) no longer see this speed halving issue for multiple transfers over SMB.

• with AIO enabled, HDDs in the array don't perform well over SMB, and will typically be slower than with AIO disabled, even for single transfers (pausing and unpausing Windows Explorer may help, but it's not really a solution).

I've forwarded this info to LT, maybe they can do something about it in the future.

If the issues with array HDDs and Samba AIO could be resolved, they could enable it by default.

Hey JorgeB,

Sorry to jump back in on this old post but did you ever hear anything back about this issue?

It's known to LT, but I'm not sure when they will be able to look at it.

No probem, we will just have to wait and keep our fingers crossed. Thanks for all the help!