root@unraid:/mnt/cache#  sync; dd if=/dev/zero of=test.img oflag=direct bs=1M count=1024; sync; dd of=/dev/null if=test.img bs=1M count=1024; rm test.img;
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 46.4026 s, 23.1 MB/s
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 8.12212 s, 132 MB/s

 

I've recently upgraded to 6.9 and rebuilt my cache pool with 1MiB-aligned MBR. While moving data around, I really noticed the speed but initially attributed it to moving a lot of very small files. However, the speed of the mover when moving large files was also painfully slow. Writing directly to the array disks is much faster (~180 MB/s in/out) but what you'd expect from a spinning disk (and the cache SSDs are on the same controller as a couple of the array disks).

 

I'm unclear if this problem is new in 6.9 but I suspect it's not since I was noticing some performance issues previously and had taken some steps to alleviate them (offloading some appdata folders to an unassigned nvme drive).

unraid-diagnostics-20210308-1450.zip

13 minutes ago, doguitar said:

Writing directly to the array disks is much faster (~180 MB/s in/out) but what you'd expect from a spinning disk

That's considerably faster than I would expect for writes to the parity array and makes me wonder if you aren't misinterpreting something.

 

See here for an explanation of why writes to an array with parity cannot happen at the speed of "a spinning disk":

 

 

root@unraid:/mnt/disk1# sync; dd if=/dev/zero of=test.img oflag=direct bs=1M count=1024; sync; dd of=/dev/null if=test.img bs=1M count=1024; rm test.img;
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 6.57578 s, 163 MB/s
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 5.39734 s, 199 MB/s