I'm curious if I can do anything to increase my performance, especially during parity checks.

 

Ryzen 2600, 48GB of RAM running 6.9.0-beta35

All drives are 10TB Iron Wolfs, except one of the dual parity drives which is a Western Digital (WD101KFBX-68R56N0).  I'm running 1.5TB of RAID1 SSD cache.  Everything is running BTRFS.  I'm using reconstruct write, and I don't power down any of my drives.  I was never able to get that to work properly, but I'm not opposed to it.

 

SAS9211-8I 8PORT Int 6GB Sata+SAS Pcie 2.0 (flashed to IT mode, shows up as Serial Attached SCSI controller: Broadcom / LSI SAS2008 PCI-Express Fusion-MPT SAS-2 [Falcon] (rev 03)) (Running 2118it)

HP 468405-002 PCIE SAS EXPANDER CARD 468405-001 487738-001 (HP SAS EXP Card  2.10  -)

 

[0:0:0:0]	disk    SanDisk' Cruzer Fit       1.00  /dev/sda   31.4GB
[9:0:0:0]	disk    ATA      ST10000VN0004-1Z SC60  /dev/sdb   10.0TB
[9:0:1:0]	disk    ATA      Samsung SSD 860  2B6Q  /dev/sdc   1.00TB
[9:0:2:0]	disk    ATA      WDC WDS500G2B0A  90WD  /dev/sdd    500GB
[9:0:3:0]	disk    ATA      Samsung SSD 860  2B6Q  /dev/sde   1.00TB
[9:0:4:0]	disk    ATA      SanDisk SDSSDH35 70RL  /dev/sdf    500GB
[9:0:5:0]	disk    ATA      ST10000VN0004-1Z SC60  /dev/sdg   10.0TB
[9:0:6:0]	disk    ATA      ST10000VN0004-1Z SC60  /dev/sdh   10.0TB
[9:0:7:0]	disk    ATA      ST10000VN0008-2J SC60  /dev/sdi   10.0TB
[9:0:8:0]	disk    ATA      ST10000VN0004-1Z SC60  /dev/sdj   10.0TB
[9:0:9:0]	disk    ATA      ST10000VN0008-2J SC60  /dev/sdk   10.0TB
[9:0:10:0]	disk    ATA      WDC WD101KFBX-68 0A03  /dev/sdl   10.0TB

 

I was running a BUNCH of maintenance scripts, but someone told me that was a waste.  I'm not sure what type of maintenance I need of my drives.

I was running balances (weekly I believe).  I'm not doing any balance on my data drives, I was told it wasn't required.

/usr/bin/ionice --class idle /usr/bin/nice --adjustment=19 /sbin/btrfs balance start -musage=50 /mnt/disk1 > /dev/shm/disk1.balance.output
/usr/bin/ionice --class idle /usr/bin/nice --adjustment=19 /sbin/btrfs balance start -dusage=90 /mnt/disk1 >> /dev/shm/disk1.balance.output

Scrubs are running twice a month, 15 days apart...one drive a time.

/usr/bin/ionice --class idle /usr/bin/nice --adjustment=19 /sbin/btrfs scrub start -Bd -c 2 -n 5 /mnt/disk4 > /dev/shm/disk4.scrub.output

Cache maintenance looks like this, daily.

/usr/bin/ionice /usr/bin/nice /sbin/btrfs scrub start -Bd -c idle /mnt/cache > /dev/shm/cache.scrub.output

I'm also running daily smart checks at 9am.

for i in {b..o}; do
    smartctl --test=short /dev/sd$i
done

If my controller/expander are my bottleneck, what is the current suggested setup?

Expander was 3Gbps type, if you setup in single link, then some bottleneck will occur ( count on 7 disks ) , but it won't that slow. Some hidden problem not relate above info.

 

You need post diagnostics and better provide diskspeed test ( docker apps ) result.

 

57 minutes ago, Doug Eubanks said:

Scrubs are running twice a month

Suggest not scrubs on array disk, if I am correct it means fully write all data ( also parity ), I concern disk failure happen or data corrupt. Also, I don't think this have much benefit.

 

I just keep 200GB space per disk and parity check monthly.

 

From experience, slow parity check mainly because

- PCIe bandwidth or link width issue

- One or some disk have problem

- One or some disk busy with other task.

- CPU in power safe mode, clock speed too low

 

 

Edited December 1, 20205 yr by Vr2Io

Thanks for your reply.

 

I paused my parity check and ran the disk speed test.  My diagnostics are included as well.

 

Thanks,

Doug

unraid-diagnostics-20201201-1412.zip

Run also the controller benchmark and post the results, make sure the screenshot shows the link speed.

SAS2008 PCI-Express Fusion-MPT SAS-2 [Falcon]

Broadcom / LSI
Serial Attached SCSI controller

Type: Onboard Controller
Current & Maximum Link Speed: 5GT/s width x8 (4 GB/s max throughput)
Capabilities: storage pm pciexpress vpd msi msix bus_master cap_list rom

 

While the controller is bottlenecking a little it's not the issue, please post diags during a parity check, make sure no other array activity is going on, there were some small writes to the array in the previous diags.

I stopped all of my VMs and all of my dockers.  I ran the diags and benchmarks while the system is completely idle.  I'm still trying to complete the benchmarks while parity is running, but I keep getting "speed gap" warnings and retries even though I checked the box to ignore speed gaps.

unraid-diagnostics-completely idle.zip

I was unable to complete the benchmarks on parity drive 2, so I canceled the parity check.

10 hours ago, Doug Eubanks said:

I'm still trying to complete the benchmarks while parity is running

Why? That won't give accurate results.