[NBLUG/talk] ext3 performance

[Scott Doty]

On Sun, Nov 23, 2003 at 09:08:46AM -0800, Mark Street wrote:
> Keep us up on your efforts.

Out of frustration, I mounted /data and /backup (the devices used in the
impactful "tar" process) as ext2.  This actually made matters worse.

My buddy in the other list thinks there's something else going on, and I
agree.  I rolled up my sleeves and looked at hdparm:

# hdparm /dev/hdb

/dev/hdb:
 multcount    = 16 (on)
 IO_support   =  0 (default 16-bit)
 unmaskirq    =  0 (off)
 using_dma    =  0 (off)
 keepsettings =  0 (off)
 readonly     =  0 (off)
 readahead    =  0 (off)
 geometry     = 30401/255/63, sectors = 488397168, start = 0

(Same for /dev/hda...only difference from the defaults is I've turned off
readahead.)

Something that really bugs me:
 _ _ _
# hdparm -d1 /dev/hdb

/dev/hdb:
 setting using_dma to 1 (on)
 HDIO_SET_DMA failed: Operation not permitted
 using_dma    =  0 (off)
 _ _ _

I should explain:  this is Intel's onboard SATA chipset.  Unfortunately,
Linux doesn't support the chipset in "native" mode (which is a pain, I
wanted to use the onboard RAID-1).  So it's currently running in "legacy"
mode, which seems to have seriously degraded performance.

However, I just tried this on both drives:

 _ _ _
# hdparm -c1 /dev/hda

/dev/hda:
 setting 32-bit IO_support flag to 1
 IO_support   =  1 (32-bit)
 _ _ _

Doesn't seem to have made much of a difference, nor did flipping the
(un)mask interrupt bit.

I think my next move is to get the thing running in "native" mode.

 -Scott