very weird: entire system "blocks" during hard-drive IO

very weird: entire system "blocks" during hard-drive IO

Post by James Hous » Thu, 27 Jun 2002 06:06:15



Hi,

I just built a new system with the following specs:

* Asus A7N266-C motherboard (with nForce audio, but not the network or
video options)

* Athlon XP 18800+ (1.53GHz) processor

* WD 40GB Hard Drive (about 8 months old)

* 768 MB DDR RAM

I could not get redhat 7.3 to install - I tried 5 or 6 times and each
time it died during installation of the packages with a python error
that said something about "device busy"

I then installed rh 7.2 (no problem) and then was able to install rh 7.3
using the "upgrade" feature.

I've ended up with the following kernel version (from "uname -a"):

Linux titanium.part.net 2.4.18-3 #1 Thu Apr 18 07:32:41 EDT 2002 i686
unknown

And have KDE 3.0 as my desktop.

Every thing seems great -- except one problem... the system seems to
"block" during i/o to my hard drive. It's even a little weirder than
that, here's a description of a common occurance:

In a shell (within Konsole) I use the "gunzip" command to uncompress a
fairly large (60 megabyte) file...

After hitting return, the command runs a 5 or 10 seconds and then gives
me a new shell prompt.  During this time, every thing is fine.

Immediately following the command "completing" the entire system blocks
for another 10 seconds (cannot move mouse, or type, or anything) --
interestingly enough the "hard-drive active led" on the computer shows
that the hd is very busy during this time. To make things even weirder,
just a moment before the system "recovers", the CPU usage jumps to 100%
for about half a second or so.

Once the "block" recovers, everything is fine.  The system seems
completely usable / stable otherwise.

Any ideas???

Thanks,
James

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by Shyamal Prasa » Thu, 27 Jun 2002 08:44:44



    James> Immediately following the command "completing" the entire
    James> system blocks for another 10 seconds (cannot move mouse, or
    James> type, or anything) -- interestingly enough the "hard-drive
    James> active led" on the computer shows that the hd is very busy
    James> during this time. To make things even weirder, just a
    James> moment before the system "recovers", the CPU usage jumps to
    James> 100% for about half a second or so.

Is DMA enabled on your drive? Use hdparm to check.

/Shyamal

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by James Hous » Thu, 27 Jun 2002 09:09:54


That did it ... Thanks!   (I learn something new every day!)

BTW: Is there a standard way to make the setting "permanent" or do I
just add it to a start-up script?

James



>     James> Immediately following the command "completing" the entire
>     James> system blocks for another 10 seconds (cannot move mouse, or
>     James> type, or anything) -- interestingly enough the "hard-drive
>     James> active led" on the computer shows that the hd is very busy
>     James> during this time. To make things even weirder, just a
>     James> moment before the system "recovers", the CPU usage jumps to
>     James> 100% for about half a second or so.

> Is DMA enabled on your drive? Use hdparm to check.

> /Shyamal

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by Skylar Thompso » Thu, 27 Jun 2002 14:46:53



> That did it ... Thanks!   (I learn something new every day!)

> BTW: Is there a standard way to make the setting "permanent" or do I
> just add it to a start-up script?

You can either add it to /etc/rc.local, or you can edit
/etc/sysconfig/harddisks and set the variables in there to the settings you
need.

--

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by Shya » Thu, 27 Jun 2002 15:46:25



> That did it ... Thanks!   (I learn something new every day!)

> BTW: Is there a standard way to make the setting "permanent" or do I
> just add it to a start-up script?

> James

If DMA is not enabled by default, there's usually some reason for doing
so. Search newsgroups posts for reports on stability problems for you
particular motherboard. I learnt the hard way.

One good way to stress test this is to use a shell script that keeps
doing large disk transfers, with a command like "dd if=/dev/hda
of=/dev/null", or use some stress testing program for the io subsystem.
There's one on freshmeat, cannot recall the name right now.

Shyam.

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by Shyamal Prasa » Fri, 28 Jun 2002 08:49:51



    James> That did it ... Thanks!  (I learn something new every day!)

I'm glad it worked out!

    James> BTW: Is there a standard way to make the setting
    James> "permanent" or do I just add it to a start-up script?

It depends on your distribution. Usually you add something to
/etc/rc.local or somewhere in /etc/init.d/* where it looks like it
will "look right". For example, on Slackware it used to be
/etc/rc.local as far as I remember. On Debian the best bet is to
install the hwtools package and add those lines in
/etc/init.d/hwtools.

On some distributions you want to check why DMA was not turned on,
because they try pretty hard to get this right in the first place. I
can't help you there though....

Cheers!
Shyamal

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by Francesc Guas » Fri, 12 Jul 2002 18:45:23


I have a linux server that does the same blocking when I/O.
I check the DMA using hdparm and it's ok.
If I switch it off doing hdparm -d0 /dev/hda it gets worse.

If it's turned on I get blocks but not as usually.

any ideas ?

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by Dances With Crow » Sat, 13 Jul 2002 08:13:18


On 11 Jul 2002 02:45:23 -0700, Francesc Guasch staggered into the Black
Sun and said:

Quote:> I have a linux server that does the same blocking when I/O.
> I check the DMA using hdparm and it's ok.
> If I switch it off doing hdparm -d0 /dev/hda it gets worse.

> If it's turned on I get blocks but not as usually.
> any ideas ?

Do you mean, "When 1 process is accessing the disk for extended periods
of time, all the other processes stop or slow down dramatically"?  If
that's it, you should either switch from IDE to SCSI disks, or make sure
that you've applied the -u1 setting to hdparm.  -u1 typically improves
interactive performance on IDE disks.

--
Matt G|There is no Darkness in Eternity/But only Light too dim for us to see
Brainbench MVP for Linux Admin /
http://www.brainbench.com     /  "He is a rhythmic movement of the
-----------------------------/    penguins, is Tux." --MegaHAL

 
 
 

very weird: entire system "blocks" during hard-drive IO

Post by Francesc Guas » Tue, 16 Jul 2002 17:55:33



Quote:> On 11 Jul 2002 02:45:23 -0700, Francesc Guasch staggered into the Black
> Sun and said:
> > I have a linux server that does the same blocking when I/O.
> Do you mean, "When 1 process is accessing the disk for extended periods

No. It happens sometimes when I do cd, ls I edit a small file and so on.

Quote:> of time, all the other processes stop or slow down dramatically"?  If

What I notice is that many times, these commands take some
seconds to answer and should have answered inmediately.

Quote:> that's it, you should either switch from IDE to SCSI disks, or make sure
> that you've applied the -u1 setting to hdparm.  -u1 typically improves
> interactive performance on IDE disks.

 Timing buffer-cache reads:   128 MB in  0.97 seconds =131.96 MB/sec
 Timing buffered disk reads:  64 MB in  2.28 seconds = 28.07 MB/sec

this is the info of the drive:

# hdparm -i /dev/hda

/dev/hda:

 Model=ST320410A, FwRev=3.34, SerialNo=3FG092G7
 Config={ HardSect NotMFM HdSw>15uSec Fixed DTR>10Mbs RotSpdTol>.5% }
 RawCHS=16383/16/63, TrkSize=0, SectSize=0, ECCbytes=4
 BuffType=unknown, BuffSize=2048kB, MaxMultSect=16, MultSect=16
 CurCHS=16383/16/63, CurSects=16514064, LBA=yes, LBAsects=39102336
 IORDY=on/off, tPIO={min:240,w/IORDY:120}, tDMA={min:120,rec:120}
 PIO modes: pio0 pio1 pio2 pio3 pio4
 DMA modes: mdma0 mdma1 mdma2 udma0 udma1 udma2 udma3 *udma4 udma5
 AdvancedPM=yes: unknown setting WriteCache=enabled
 Drive Supports : Reserved : ATA-1 ATA-2 ATA-3 ATA-4 ATA-5 ATA-6

 
 
 

1. weird problem: system "blocks" during io

Hi,

I just built a new system with the following specs:

* Asus A7N266-C motherboard (with nForce audio, but not the network or
video options)

* Athlon XP 18800+ (1.53GHz) processor

* WD 40GB Hard Drive (about 8 months old)

* 768 MB DDR RAM

I could not get redhat 7.3 to install - I tried 5 or 6 times and each
time it died during installation of the packages with a python error
that said something about "device busy"

I then installed rh 7.2 (no problem) and then was able to install rh 7.3
using the "upgrade" feature.

I've ended up with the following kernel version (from "uname -a"):

Linux titanium.part.net 2.4.18-3 #1 Thu Apr 18 07:32:41 EDT 2002 i686
unknown

And have KDE 3.0 as my desktop.

Every thing seems great -- except one problem... the system seems to
"block" during i/o to my hard drive. It's even a little weirder than
that, here's a description of a common occurance:

In a shell (within Konsole) I use the "gunzip" command to uncompress a
fairly large (60 megabyte) file...

After hitting return, the command runs a 5 or 10 seconds and then gives
me a new shell prompt.  During this time, every thing is fine.

Immediately following the command "completing" the entire system blocks
for another 10 seconds (cannot move mouse, or type, or anything) --
interestingly enough the "hard-drive active led" on the computer shows
that the hd is very busy during this time. To make things even weirder,
just a moment before the system "recovers", the CPU usage jumps to 100%
for about half a second or so.

Once the "block" recovers, everything is fine.  The system seems
completely usable / stable otherwise.

Any ideas???

Thanks,
James

2. NCSA on Linux prob (different one) trailing backslash??

3. strange problems rh 7.3: system "blocks" during io

4. Are thin clients the wave of the future?

5. "640MB optical drive" or "Support for block size > 1K"

6. dialing in to Linux

7. GETSERVBYNAME()????????????????????"""""""""""""

8. flushing Directory Name Lookup Cache (DNLC) on Solaris 2.4

9. How to prevent "apm" from turning off hard drive during uptime?

10. Hard drive access during "idle time"

11. "boot0: error80" during hard drive boot process

12. During Linux install "Disk Druid" not recognizing 20GB Hard Drive properly....

13. """"""""My SoundBlast 16 pnp isn't up yet""""""""""""