I've posted about disk throughput before, and I have seen a number of other posts about it as well. So, I'm back with another question, hoping to either come to an understanding of why it works this way or potentially expose an underlying issue.

I have two Samsung 870 EVO drives in a ZFS mirror as a cache drive for a shared folder. Everything written to the share goes to the cache (obviously). My network backbone is 10G, although the UR server utilizes the MB's built-in 2.5G NIC.

I was transferring a folder with a series of +12GB files over to the array, and the network transfer rate was about 30-40MB/s. Which is a far cry from 10G speeds. That's when I took a look at the UR dashboard to see transfer rates and disk throughput.

The network throughput was showing approximately 600 Mb/s, and disk throughput was fluctuating between 0 and 170 MB/s. There was never a constant write speed to the cache drive for the shared folder.

The mover is not running, and no other transfers were running in the background.

If there are any tests I can run, please let me know.

Diag attached.

Thx,

A.

Edited July 22, 2025Jul 22 by aglyons

28 minutes ago, aglyons said:

was transferring a folder with a series of +12GB files over to the array, and the network transfer rate was about 30-40MB/s. Which is a far cry from 10G speeds.

You will never get close to 10GbE transferring to a spinning rust array; you can still get better speeds with turbo write enabled, assuming no controller bottlenecks.

You will never get close to 10GbE transferring to a spinning rust array; you can still get better speeds with turbo write enabled, assuming no controller bottlenecks.

I did state that the files are being transferred to an SSD ZFS mirror cache drive.

2025-07-21_10-02-35.mkv 2025-07-21_10-02-04.mkv

39 minutes ago, aglyons said:

I was transferring a folder with a series of +12GB files over to the array,

If by array you mean a pool, please use that term; array typically means "the array."

If you were getting 30 30MB/s transferring to the pool that is very slow. Start by running an iperf test to confirm the LAN is OK, then you can test the devices.

I'll run the iperf.

Just so I am clear, if an array folder has a cache-first config, when writing to the array, the files will initially be copied onto the cache pool, and then the mover 'moves' the files to the array. If this is not the 'understood' function, do I have to specifically state the files are landing on a cache pool, regardless of the fact that the mover will relocate the files?

I chose to be specific about the config in the event that a pool array behaves differently than a shared folder cache config. Yes, they both land on a pool device, but in the latter, the mover is involved, which may introduce other parameters.

It will be the same if it's actually writing to the pool, it's just to make things clear.