Dear all,
I have two identical DS20s, one of which runs its database off a pretty
external RAID array (RA3000) via a FWD card and the other, which is
about to receive a similar RAID setup, is currently running off the
internal disk shelf and the standard FW SCSI card. Machines are
running 4.0F PK#4.
The tuning on both boxes is identical, including smsync2,throttle on
the data disks:
/dev/rz2h /data ufs rw,smsync2,throttle 1 3
/dev/rz3h /data_backup ufs rw,smsync2,throttle 1 3
vfs:
io-throttle-maxmzthroughput = 1
io-throttle-shift = 2
smoothsync-age = 60
The disks are standard DEC-issue FW 9Gb disks:
rz2 at scsi0 target 2 lun 0 (LID=1) (COMPAQ BB00912301 B016)
(Wide16)
rz3 at scsi0 target 3 lun 0 (LID=2) (COMPAQ BB00912301 B016)
(Wide16)
The problem is that a "vacuum" on the PostgreSQL database on the
machine with the RAID array takes 5 minutes against almost 50 minutes
on the other. I do appreciate that RAID arrays are fast but this much
faster?
Checking the running database shows that it is sitting there waiting
on read() and iostat doesn't show very impressive figures (in the mid-
500s under the bps column). I have seen the disk do high 3000s in the
bps column while extracting a tar file or doing other disk I/O so I am
slightly confused - perhaps PostgreSQL is being very inefficient in
accessing the data?
Any pointers welcome.
Arrigo
Received on Fri Aug 04 2000 - 11:20:33 NZST