System slowdown on large disk writes

Walter Dnes waltdnes-SLHPyeZ9y/tg9hUCZPvPmw at public.gmane.org
Sat Mar 4 06:13:18 UTC 2006


On Thu, Mar 02, 2006 at 04:25:16PM -0500, William O'Higgins Witteman wrote
> On Thu, Mar 02, 2006 at 03:03:51PM -0500, Lennart Sorensen wrote:
> >On Thu, Mar 02, 2006 at 01:17:50PM -0500, William O'Higgins Witteman wrote:
> >> I'm copying 500Mb to 4Gb files to another location on the same ext3
> >> (I think - how would I check?) 140-odd Gb partition.
> >
> >I believe the default io schedular in 2.6 kernels is designed to handle
> >random io pretty nicely, but can be pretty unfair if doing lots of
> >sequential I/O.  Perhaps trying another io scheduler, or tuning it a bit
> >(it should have options in /sys or /proc somewhere) could make it
> >better.
> 
> Eeek.  Okay, I've never poked into that part of the filesystem before.
> I get to learn something new.

  I think the problem here is ext3, which is actually a glorified ext2
with some journaling overhead added.  First, run hdparm and see if you
really do have DMA enabled.  If you're *REALLY* adventurous "man hdparm"
says...

-u     Get/set interrupt-unmask flag for the drive.   A  setting  of  1
       permits  the driver to unmask other interrupts during processing
       of a disk interrupt, which greatly improves Linux's  responsive-
       ness and eliminates "serial port overrun" errors.  Use this fea-
       ture with caution: some  drive/controller  combinations  do  not
       tolerate  the increased I/O latencies possible when this feature
       is enabled, resulting in massive filesystem corruption.  In par-
       ticular, CMD-640B and RZ1000 (E)IDE interfaces can be unreliable
       (due to a hardware flaw) when this option is  used  with  kernel
       versions  earlier  than 2.0.13.  Disabling the IDE prefetch fea-
       ture of these interfaces (usually a BIOS/CMOS setting)  provides
       a safe fix for the problem for use with earlier kernels.

-- 
Walter Dnes <waltdnes-SLHPyeZ9y/tg9hUCZPvPmw at public.gmane.org> In linux /sbin/init is Job #1
My musings on technology and security at http://tech_sec.blog.ca
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml





More information about the Legacy mailing list