[MacDV] Re: more about the inutility of defragmenting an OS X FS.

[Peter van der Linden]

On Dec 31, 2003, at 11:02 AM, Mark M. Florida wrote:

> It's certainly true that modern drive and OS technology lessens the 
> need for defragging, but that's for what would be considered "normal" 
> usage like e-mail, word processing, and even editing small images.

Neither the kernel, nor the disk i/o subsystem has any idea what 
applications are being run. They just see a stream of I/O requests.  
Most applications are i/o bound (video encoding is an exception, but 
video encoding is not a real time constrained operation), so most 
applications look the same to the OS and i/o subsystem. I.e. disk i/o 
requests come in asynchronously, and are serviced asynchronously.

You don't and can't know how the blocks are laid out on physical disk.  
You can't know it  because that information doesn't leave the disk 
controller.  The controller maintains a fiction of putting the blocks 
where you ask, but it actually puts the blocks wherever it wants to.  
(And the algorithms that the controller uses in its attempts to 
optimize logical block placement are highly guarded secrets, too). The 
nice GUI maps can only show you the fiction, not the physical 
on-platter reality.  For that reason alone, rearranging disk blocks to 
make the map look pretty, does nothing for performance.

On top of that, you have all the buffering and cacheing done by the 
kernel I/O subsystem.  And on top of that, you have unrelated disk 
accesses done by the OS, such as swap, tempfs, journalling, and i/o for 
other processes.  These are going to put the disk heads wherever they 
want, and attempts to make files contiguously allocated therefore yield 
insignificant performance improvements except possibly under 
pathological conditions.

A separate drive for video is a good idea, but because of dedicated 
performance, not because of fragmentation.  If you're a true believer 
in defragmentation then putting your video on a separate partition on 
the same drive is a terrible idea - it guarantees that all other 
references to disk will reposition the heads.

There's a good layman's intro to modern disk technology at
http://www.acmqueue.org/modules.php?name=Content&pa=showpage&pid=46

But, hey, prove me wrong, show me some data that supports claims that 
disk defragmentation improves application performance in MacOS 10.3  
Tell me how I can reproduce this alleged performance drop on one of my 
own systems.

      Peter