Re: Ext3fs performance/kjournald deadlock issue

[Edward Moon <em mooned org>]

I ran hdparm and this is what I get:
	/dev/sda1:
	Timing buffered disk reads: 64MB in 4.11 seconds = 15.57 MB/sec

	/dev/hda1:
	Timing buffered disk reads: 64MB in 708.04 seconds = 92.56 kB/sec

Hmm. I checked /var/log/messages and found:
Dec 30 20:16:42 ex kernel: hda: drive not ready for command
Dec 30 20:17:41 ex kernel: hda: timeout waiting for DMA
Dec 30 20:17:41 ex kernel: hda: status error: status=0x58 { DriveReady
SeekComplete DataRequest }
Dec 30 20:17:41 ex kernel: hda: drive not ready for command
Dec 30 20:18:22 ex kernel: hda: timeout waiting for DMA
Dec 30 20:18:22 ex kernel: hda: status error: status=0x58 { DriveReady
SeekComplete DataRequest }
Dec 30 20:18:22 ex kernel: hda: drive not ready for command
Dec 30 20:18:53 ex kernel: hda: timeout waiting for DMA
Dec 30 20:18:53 ex kernel: hda: status error: status=0x58 { DriveReady
SeekComplete DataRequest }
Dec 30 20:18:53 ex kernel: hda: drive not ready for command
Dec 30 20:21:57 ex kernel: hda: irq timeout: status=0xd0 { Busy }

Checking hdparm I see:

/dev/hda:
 multcount    =  0 (off)
 I/O support  =  0 (default 16-bit)
 unmaskirq    =  0 (off)
 using_dma    =  0 (off)
 keepsettings =  0 (off)
 nowerr       =  0 (off)
 readonly     =  0 (off)
 readahead    =  8 (on)
 geometry     = 4982/255/63, sectors = 80041248, start = 0

I'm going to play around with hdparm settings and checkout options in the
BIOS.

Thanks for the help!
On Sun, 30 Dec 2001, Andrew Morton wrote:

> Edward Moon wrote:
> >
> > Hi,
> >
> > I've got a Redhat 7.2 system:
> > Celeron 400
> > 512MB RAM
> > 18GB SCSI HD (/boot (ext2fs), / (ext3fs))
> > 40GB IDE HD (/ (ext3fs))
> >
> > I run a bunch of services on the box (apache + mod_perl, MYSQL, Samba,
> > etc) but the system is not under heavy load.
> >
> > This system has performance issues reading/writing to the ext3fs
> > filesystems. The performance issues cropped up when I installed Redhat
> > 7.2. Previously the system ran Redhat 6.2 and didn't have the following
> > issues:
> >
> > Copying files from the IDE drive to the SCSI drive is very
> > slow (over 30 minutes transferring 50MB of data).
>
> That's ext3 to ext3, yes?
>
> Something is certainly grossly wrong.  Have you tried mounting all
> partitions as ext2?  That'll tell us if it's a filesystem problem,
> or something lower down.
>
> You should run `hdparm -t /dev/hdXX' and `hdparm -t /dev/sdXX' to
> verify that the disk is running at the expected raw speed.  I suspect
> that they are not.   Possibly the 7.2 IDE driver decided to run
> your disk in some very slow PIO mode.   You can experiment with
> hdparm parameters to tune this up, but be careful - it's easy to
> lock your machine when doing this.
>
> > Additionally, I can't play mp3 files from a samba share on this
> > machine without winamp pausing every 5-10 seconds. If I copy the mp3 files
> > to the local hard drive or to a samba share on a Redhat 6.2 system
> > (running ext2fs), the files will play fine.
> >
> > I have noticed that cp/smbd seems to deadlock with kjournald. I haven't
> > modifed any of the ext3fs settings.
>
> A complete deadlock?  What is it which indicates that the lockup
> is in kjournald?
>
> > Can I improve performance by changing the journal settings? I think the
> > default is to update the journal every 5 seconds? If I increase that to 60
> > seconds would that improve performance?
> >
>
> It won't help.  First thing to check is that the raw disk throughput
> is decent.
>
> -
>