When disk IO goes bad

[Gregory Boyce]

On Thu, 2004-02-05 at 09:26, Duane Morin wrote:
> My server lately has been dog slow.  I assumed it had to do with web 
> server problems since all I really run is Tomcat and smtp.  Well, I
> turned off Tomcat for now.  
> 
> The weird thing is that there's basically no load on the machine -- about 
> a 0.09 on average.  BUT, whenever I do anything file system related, it 
> shoots through the roof.  For example just this morning I copied a 12meg 
> file from one directory to another, and the load shot up to 8.0.
> 
> What in the world causes THAT?  I mean, sure, it's not the newest hard 
> drive in the world, but it never did that before.  Can a drive begin to 
> die in such a way that it starts to put more load on the machine?  That 
> seems pretty weird to me.  I'm more likely to believe that I've just 
> filled up the drive with too many individual files and am now running into 
> some sort of inode problem or something.

Generally it means that DMA is not enabled on the hard drive.

Run "hdparm /dev/hda" (assuming hda is your hard drive here).

You should see something like:

/dev/hda:
 multcount    = 16 (on)
 IO_support   =  1 (32-bit)
 unmaskirq    =  1 (on)
 using_dma    =  1 (on)
 keepsettings =  0 (off)
 readonly     =  0 (off)
 readahead    = 256 (on)
 geometry     = 65535/16/63, sectors = 234375000, start = 0

If using_dma is not 1, try enabling it with "hdparm -d 1 /dev/hda".  If
that does not work, then your drive or IDE controller might not support
DMA, or you might not have DMA enabled in your kernel.

Are you running a kernel from a major distro, or something custom?