Copying between user-shares are at about 15 to 35 MB/s (no parity, no cache) - Client directly to user-share is fine.

[enJOyIT]

Hi,

 

when I'm copying between one user-share (only on one disk) to another user-share (splitted over serveral disks) I get only about 15 to 35 MB/s.

 

Tried with Windows (SMB) and Midnight Commander directly on the server.

 

 

 

When I am copying directly from my Windows machine to the user-share I get full speed (around 160 MB/s as the Windows machine is an VM with virtio network driver).

 

Reading speed from source user-share is also fine.

 

Why does unraid is doing this?

Edited January 22, 2022 by enJOyIT

[Squid]

Post your diagnostics.  Also, does the source and destination wind up overlapping? (ie: the same disk winds up being involved?)

 

What is the number of files involved here in this copy?  A couple large ones or tons of very small files?

[enJOyIT]

It is one big file (around 2 GB) and there are preclears of other disks going on... if this is important.

 

There are two different disks involved.

 

I can deliver diagnostic information tomorrow (I am from germany). sorry for the delay.

Edited January 22, 2022 by enJOyIT

[Squid]

Preclears could be possibly maxing out the available bandwidth to the controllers (diagnostics would help on that).  

 

The writes from via SMB would wind up being cached in memory and slowly actually written to the disk.  Direct copies from share to share I would think should do the same, but there's also a lot of variables to consider.  And the multiple preclears happening simultaneously is my first suspect without seeing the diagnostics

[enJOyIT]

Well, thank you for your fast reply!

I'll check the transfer speed again after the preclears are finished and will give a feedback and possibly a diagnostic log if the issue is still there.

[enJOyIT]

Ok, the problem was related to the ongoing preclearing (writing zeros) of a bunch of disks. After they were finished the copyspeed is at the expected speed. 🙂