Re: Extremely poor disk performance on ProLiant DL360 G5 RAID

[Frank Wille <frank%phoenix.owl.de@localhost>]

On Thu, 19 Jul 2018 14:26:38 +0000
maya%netbsd.org@localhost wrote:

> On Thu, Jul 19, 2018 at 12:03:29PM +0200, Frank Wille wrote:
> > ...and two 550 GB SAS RAID-1 disks, configured via the BIOS and appearing
> > as a single SCSI disk:
> > ciss0 at pci7 dev 0 function 0: HP Smart Array 3
> > ciss0: interrupting at ioapic0 pin 16
> > ciss0: 1 LD, HW rev 3, FW 5.20/5.20, 64bit fifo
> > scsibus0 at ciss0: 1 target, 1 lun per target
> 
> This might be relevant:
> https://v4.freshbsd.org/commit/netbsd/src/a5QCd7STkggvNCqA

Ok, I have built a new kernel with "options CISS_NO_INTERRUPT_HACK", but
as expected that only makes things worse.

Benchmark of extracting the 8.0 base.tgz set, with the original GENERIC
kernel:
      149.79 real         1.74 user         2.23 sys

And with a CISS_NO_INTERRUPT_HACK kernel:
      277.11 real         1.52 user         2.11 sys

For comparison, the old server (Supermicro PDSM4+) with Adaptec RAID,
Xeon 2.13 GHz, running NetBSD 6.1.5:

aac0 at pci5 dev 14 function 0: Adaptec RAID 3405
aac0: interrupting at ioapic0 pin 18
aac0: Enabling 64-bit address support
aac0: Enable 64-bit array support
aac0: New comm. interface enabled
aac0: XScale 80321 at 500MHz, 128MB mem (111MB cache), optional battery not installed
aac0: Kernel 5.2-0 [Build 15323], Monitor 5.2-0 [Build 15323], S/N 12CB1D
aac0: Controller supports: 0x7f1d7d<SNAPSHOT,WCACHE,DATA64,HOSTTIME,RAID50,WINDOW4GB,SOFTERR,SGMAP64,ALARM,NONDASD>
ld0 at aac0 unit 0: RAID 1 (Mirror)
ld0: 464 GB, 60700 cyl, 255 head, 63 sec, 512 bytes/sect x 975155200 sectors
ppb5 at pci4 dev 0 function 2: Intel IOP333 PCI Express-to-PCI Bridge #1 (rev. 0x00)

Same base.tgz extraction:
real    0m2.854s
user    0m1.545s
sys     0m1.265s

That's less than 3 seconds, compared to 150!

-- 
Frank Wille