NetBSD-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

poor raid5 performance with mfi0

I have an ix systems box with a smc-rebranded lsi logic controller:

mfi0 at pci3 dev 0 function 0mfi0: interrupting at ioapic1 pin 2
mfi0: logical drives 2, version 12.12.0-0047, 512MB RAM
scsibus0 at mfi0: 64 targets, 8 luns per target
scsibus0: waiting 2 seconds for devices to settle...
sd0 at scsibus0 target 0 lun 0: <SMC, SMC2108, 2.12> disk fixed
sd0: fabricating a geometry
sd0: 237 GB, 243186 cyl, 64 head, 32 sec, 512 bytes/sect x 498044928 sectors
sd0: fabricating a geometry
sd1 at scsibus0 target 1 lun 0: <SMC, SMC2108, 2.12> disk fixed
sd1: fabricating a geometry
sd1: 3721 GB, 3810880 cyl, 64 head, 32 sec, 512 bytes/sect x 7804682240 sectors
sd1: fabricating a geometry

#> bioctl mfi0 show
Volume Status       Size         Device/Label    Level Stripe
     0 Online       237G                mfi0    RAID 0    N/A  65535 seconds
   0:0 Online       238G         1:0.0 noencl <ATA     C300-CTFDDAC256M0006>
     1 Online       3.6T                mfi0    RAID 5    N/A  65535 seconds
   1:0 Online       932G         1:1.0 noencl <ATA     WDC WD1003FBYX-01V01>
   1:1 Online       932G         1:2.0 noencl <ATA     WDC WD1003FBYX-01V01>
   1:2 Online       932G         1:3.0 noencl <ATA     WDC WD1003FBYX-01V01>
   1:3 Online       932G         1:4.0 noencl <ATA     WDC WD1003FBYX-01V01>
   1:4 Online       932G         1:5.0 noencl <ATA     WDC WD1003FBYX-01V01>

sd0, passthrough of an ssd works great.

sd1, which should be raid5 of 5 fast (real) disks, works nicely on read,
over 500 MB/s:

#> dd if=/dev/rsd1d of=/dev/null count=8192 bs=256k
8192+0 records in
8192+0 records out
2147483648 bytes transferred in 4.183 secs (513383611 bytes/sec)

But writing is atrocious at 7 MB/s:

#> dd of=/dev/rsd1d if=/dev/zero count=256 bs=256k
256+0 records in
256+0 records out
67108864 bytes transferred in 9.873 secs (6797210 bytes/sec)

FFS performance is similarly bad, with or without wapbl, surely
dominated by the poor write performance.

So, I'm going to do RAID10 and a hot spare, which will hopefully get me
1.8T that's reasonable.

Any words of wisdom to do something else?

Is the controller junky, or is our driver deficient, or is it likely our
stripe size is wrong (not sure what it is; someone else set it up and
bioctl doesn't show it, but surely it's default from the bios setup)?

Attachment: pgpNxIC7BveqF.pgp
Description: PGP signature

Home | Main Index | Thread Index | Old Index