10 minutes ago, StatMatt said:

and maybe ZFS just has too much overheard for a 11-year-old CPU?

It's not really ZFS overhead, it's an Unraid issue that hopefully will be fixed for v6.13, but yes, XFS (or btrfs) should perform noticeably better on the same drives, if they are assigned to the array, no issue with ZFS when assigned to pools.

I am still facing the issue where read speed from the server is weird. These 2 file is located in the same share same disk (disk6). Docker and VM is disabled to ensure nothing is affect the result. both transfer is a single large file. I repeated this test and keep getting this result. How does data from the same drive same share transfer at different speed over the network?

 

 

Transferring 20+GB file to my workstation desktop which is on NVME

 

 

Transferring 40+GB file to my workstation desktop which is on NVME

tower-diagnostics-20230831-0016.zip

I guess I have finally found out the reason why this is happen to me.  Fragmentation. I am using Deluge as my torrent client and I am using it heavily to download huge torrent. There is a setting call "Pre-allocate disk space" is left uncheck all these while. This result in those huge single file is not store sequential on the hard disk. This impact the read performance. I have put this hypothesis to the test and the result is promising. I download the torrent twice with "Pre-allocate disk space" enabled and disabled. One first test it is enabled and once the torrent downloaded finish. I remove the job from the client ensure that nothing is accessing the file. I was able to achieve a read speed of 200+MB/s for the while transfer of 50GB. I delete everything and move on to the second test. I disable the "Pre-allocate disk space" and download the same torrent once again. This time round my read speed hovering at 100+MB/s. In another word, the option "Pre-allocate disk space" does help to ensure that data is written in sequential manner.