Horrendous RAIDframe reconstruction performance

[John Klos <john%ziaspace.com@localhost>]

Hello,

I'm setting up two helium, non-SMR, 512e 8 TB disks (HGST HUH728080ALE604) 
in a RAIDframe mirror:

[     2.829768] wd2 at atabus2 drive 0
[     2.829768] wd2: <HGST HUH728080ALE604>
[     2.829768] wd2: drive supports 16-sector PIO transfers, LBA48 addressing
[     2.829768] wd2: 7452 GB, 15504021 cyl, 16 head, 63 sec, 512 bytes/sect x 15628053168 sectors (0 bytes/physsect; first aligned sector: 8)
[     2.859768] wd2: GPT GUID: 4086e8f6-0ddd-4689-a942-80bf1b598539
[     2.859768] dk0 at wd2: "raid8tb0", 15611274240 blocks at 1024, type: raidframe
[     2.859768] dk1 at wd2: "swap8tb0", 16777216 blocks at 15611275264, type: swap
[     2.869768] wd2: drive supports PIO mode 4, DMA mode 2, Ultra-DMA mode 6 (Ultra/133), WRITE DMA FUA, NCQ (32 tags) w/PRIO
[     2.869768] wd2(ahcisata0:4:0): using PIO mode 4, DMA mode 2, Ultra-DMA mode 6 (Ultra/133) (using DMA), NCQ (31 tags) w/PRIO

(Strange that it says "0 bytes/physsect")


First, it seems that autoconfigure doesn't allow forcing the root 
filesystem. I'm guessing because this is using GPT:

   Autoconfig: Yes
   Root partition: Force

Next, raidctl doesn't handle NAME= for device yet:

raidctl -v -a NAME=raid8tb1 raid0
raidctl: ioctl (RAIDFRAME_ADD_HOT_SPARE) failed: No such file or directory

Finally, even though these are absolutely not SMR drives, rebuild is 
running at 3.5 to 4 MB/sec, whether attached via USB-3 or directly 
attached via SATA:

# raidctl -v -S raid0
Reconstruction is 0% complete.
Parity Re-write is 100% complete.
Copyback is 100% complete.
Reconstruction status:
  0% |                                       | ETA: 485:13:54 -

Interestingly, a bonnie++ run shows 80+ megabytes per second in block 
writes and 160+ megabytes per second on block read, and 185 random seeks 
per second, while reconstructing.

Any thoughts about what's going on here? Is this because the drives are 
512e drives? Three weeks is a LONG time to reconstruct.

This is observed on NetBSD 9.99.68 on both i386 and on aarch64.

Thanks,
John Klos