Re: Interesting speed benchmarks

On Wed, 2007-Jan-31 07:54:24 +0200, Ian FREISLICH wrote:
>As a random datapoint that might or might not be related.  In the
>last year or so I was pontificating about how much better SCSI was
>that IDE/ATA to my boss.  I then did some benchmarks on my (CURRENT)
>system and discovered that an old 10Gig 7.5krpm UDMA33 disk
>significantly out perform some Seagate Cheatah LVD 10krpm disks.

I have access to a Sun V20z at work:
mpt0: <LSILogic 1030 Ultra4 Adapter> port 0x2000-0x20ff mem 0xfe850000-0xfe85ffff,0xfe840000-0xfe84ffff irq 27 at device 4.0 on pci2
mpt0: [GIANT-LOCKED]
mpt0: MPI Version=1.2.15.0
mpt0: Capabilities: ( RAID-1E RAID-1 SAFTE )
mpt0: 0 Active Volumes (1 Max)
mpt0: 0 Hidden Drive Members (6 Max)
da0 at mpt0 bus 0 target 0 lun 0
da0: <FUJITSU MAW3073NC 0103> Fixed Direct Access SCSI-3 device 
da0: 320.000MB/s transfers (160.000MHz, offset 127, 16bit), Tagged Queueing Enabled
da0: 70136MB (143638992 512 byte sectors: 255H 63S/T 8941C)
da1 at mpt0 bus 0 target 1 lun 0
da1: <FUJITSU MAW3073NC 0103> Fixed Direct Access SCSI-3 device 
da1: 320.000MB/s transfers (160.000MHz, offset 127, 16bit), Tagged Queueing Enabled
da1: 70136MB (143638992 512 byte sectors: 255H 63S/T 8941C)

Whilst the drives are geom mirrored, running in "prefer" mode forces
all reads to come off the first drive (iostat/systat verify this) and
I get:
# df -ki /home
Filesystem         1K-blocks      Used    Avail Capacity iused     ifree %iused  Mounted on
/dev/mirror/gm0s1f  38999124  13440472 22438724    37%  273895   4766231    5%   /home
# dump -0aL -f /dev/null -C 32 /home
  DUMP: DUMP: 13671011 tape blocks on 1 volume
  DUMP: finished in 906 seconds, throughput 15089 KBytes/sec

Note that I've only just mirrored this system that everything has been
very recently restored which might give an optimistic result but this
is still way above what I'm seeing on my ATA system at home or what
Robert is seeing.

>Last time I looked, ahc(4) was still under GIANT.  The ATA system
>has had a significant overhaul.

mpt(4) is also under giant and still gives a good account of itself.

-- 
Peter Jeremy