I'm getting weird fluctuating transfer speeds when copying from a workstation to Unraid cache drive (screenshot attached).

 

Source where I'm copying from is capable of sustaining speeds about 900MB/s, Copying from the same source to another disk attached to the workstation it will average out (~4tb folder) to about 750-900MB/s with file system overheads. I'd really like to achieve at least the same speeds when copying to unraid. 

 

The share is set up to use cache as a primary storage, and array as secondary. Cache drive itself is a btrfs raid0 with 1 nvme pcie drive and 2 sata ssd's. I'm using raid0 since I need the cache to be at least 3TB. IF this is definitely the issue I could just buy a bigger nvme drive and use it as a single drive.

 

Here are some average hdparm read test results of the individual disks, that are part of cache raid0. I haven't done any write tests yet. Is it safe to write straigth to the device when its part of raid0 with dd?

/dev/nvme0n1:
 Timing cached reads:   34890 MB in  1.98 seconds = 17630.33 MB/sec
 Timing buffered disk reads: 4242 MB in  3.00 seconds = 1413.82 MB/sec

/dev/sdc:
 Timing cached reads:   35154 MB in  1.98 seconds = 17751.61 MB/sec
 Timing buffered disk reads: 1010 MB in  3.00 seconds = 336.35 MB/sec

/dev/sdb:
 Timing cached reads:   36866 MB in  1.98 seconds = 18629.65 MB/sec
 Timing buffered disk reads: 1002 MB in  3.00 seconds = 333.70 MB/sec

 

 

Any pointers and best practices would be really appreciated what to check next and how to troubleshoot this.

 

Also if this has any significance, this is a file server / project archive for my company, not a media server. 

 

 

 

postraid-diagnostics-20240524-1321.zip

Forgot to mention that copying over SMB and from MacOS 12.7.1

Kingston SA400 have a very low sustained write performance, recommend retrying with just the NVMe device for testing, though using a Mac can also bring other issues, see here for some recommendations:

https://forums.unraid.net/topic/146818-optimal-samba-settings-to-get-better-macos-to-unraid-performance/?do=getNewComment

 

Also recommend that you make the pool an exclusive share, or use the disk share to test.

Thanks for the reply JorgeB

 

I swapped the cache raid0 to a single nvme and the issue is gone. Even without exclusive share the performance is now OK for what I need. The peaks may be a tiny bit lower but overall speed is much better, no fluctuations any more.

 

Time to get a larger nvme drive.