All

 

New Unraid user for the last 3 weeks - have 6.12.11 going - with a modest build:

 

Motherboard: ASUS ProArt Z490-Creator

CPU: Intel Core™i5 10600K (LGA1200)

MemoryCorsair Vengeance LPX CMK64GX4M2D3600C18 (64 GB)

LAN: Intel® I219V

PCIe: ASUS Hyper M.2 x16 V2

HD1 (System) Samsung USB Fit Plus (64GB) 

HD2 - Bay 1 (Parity) Seagate Exos 7E10 ST8000NM017B 8 TB

HD3 - Bay 2 (Array 1) Seagate Exos 7E10 ST8000NM017B 8 TB 

HD4 (Cache-AD) Samsung SSD 980  NVMe M2 SSD (250GB)

HD5 (Cache-VM) Samsung 970 EVO Plus NVMe M2 SSD (500GB)

HD6 (Cache-SA) Samsung 860 EVO SSD (500GB) 

HD7 (Cache-SB) Samsung 860 EVO SSD (500GB) 

 

This layout at the moment - has just a single array drive and NO Parity at all. I have about 5-6 TB of data to move from a current Windows File Server and will build out the parity after the move.

 

Network is standard 1 Gbps Ethernet, Cisco switching - stock "pro-sumer" type stuff. Testing was done from a Windows 10 22H2 ASUS workstation on the network.

 

My questions today surround my cache layout and the frustrating disk transfer speeds I am seeing. From the Unraid Main view:

 

 

I have been experimenting with the various cache combos, file systems etc and all roads seem to lead to the same end game.

 

For this real world testing - using the Unraid Shares area - I have been bouncing my Downloads share around to these different targets and sending the same test data each time:

 

1. 1 GB Testfile

2. 4.80 GB of real world data from a drive om my Windows PC. This stash contains 7 folders containing all manner of files. Big. Small. Tiny and huge.

3. AND - as a third test - I also copied this same payload - directly to a share which is ARRAY Only.

4. Finally - I copied the same fileset to the Downloads folder on my Windows 2019 server (running an old Haswell era Xeon CPU and stocked with 5900 RPM Ironwolf drives)

 

The results are both disappointing and strange

 

1. Sending the payload to ANY of the NVmE/SSD cache drives ranged from 1:10 to 1:21 to fully complete

2. Sending the same payload to the single (currently non-parity) Array drive was 1:14

3. Sending the same payload to various shares on Windows Server 2019 ranged from :57 - 1:05

 

Now - the transfer "rate" would bounce around as is expected with a wide mix of files - but everything I have read seemed to indicate that using a NVme-enabled cache (file system not withstanding) - "should" provide the fastest and most robust transfer speed possible.

 

Even more odd - my two NVmE drives listed above - are mounted within my ASUS Hyper M.2 x16 V2 PCI-e card - whose slot is basically hardwired directly to the CPU (vs being held up by the an additional chipset layer if I use the standard NMvE slots on the motherboard.) These destinations should be the fastest possible targets - but the real world tests here - show that they are no better than a rote basic EVO 860 SSD or even writing directly to the array (non-parity of course).

 

And then there are the file systems. I am experimenting with ZFS but for reasons unknown - it feels "slower" in almost all operations. I have also paired the two EVO 860's in a mirror, Raid0 and Raid1 combos and still cannot see a combo with a major advantage. Have converted to XFS and BTRFS will little upside as well.

 

All I can think of is either I have hit the theoretical maximums of the network pipe and regardless of endpoint - that is as good as it is going to get. Or something else is up with the config, file system choice, or the internals of Unraid as it sits on this specific hardware.

 

And it goes without saying that I was very disappointed to see my old creaky 2014 Windows Server run some circles around this new build. I would have thought it should be a head to head battle at least

 

Now - I am not looking to achieve 125 MB/s maximums for every file I copy. But I am seeking advice on the best combo of device/file system etc so I can complete this build, stick it in the rack and have it be reliable and modestly robust when needed. 

 

Appreciate any insights you can share.

 

S

 

Edited September 5, 20241 yr by Sonicmojo

Make sure you enable exclusive shares, that should get a you a nice performance bump, regarding the filesystems, in my experience, xfs is the fastest one, then btrfs, then zfs, but the differences are generally not very significant, so I would base my choice in the filesystem features I need/want.

1 hour ago, JorgeB said:

Make sure you enable exclusive shares, that should get a you a nice performance bump, regarding the filesystems, in my experience, xfs is the fastest one, then btrfs, then zfs, but the differences are generally not very significant, so I would base my choice in the filesystem features I need/want.

Thank you for the comments.

 

Very interesting - except for the part where Exclusive Shares cannot have a secondary storage target.

 

In my use (unlike many others it seems) Unraid here is a typical file server and the shares I want to accelerate (ideally) are those I actually send stuff to the most.

 

In this case - shares like Backup, Software see the most activity - but I have no desire to leave daily backups (Backup) or new software packages I move to the server (Software) sitting on a cache drive forever. These items need to get to the Array eventually.

 

I will take your cue and at least move back to XFS - that actually did make me feel a bit better in the middle of all these tests

 

S