Archived from groups: microsoft.public.win2000.general (
More info?)
I'm a file system guy. I have no expertise at all when it comes to hard
drive controllers/hard drives and how they work.
- Greg/Raxco Software
Microsoft MVP - Windows File System
Disclaimer: I work for Raxco Software, the maker of PerfectDisk - a
commercial defrag utility, as a systems engineer in the support department.
Want to email me? Delete ntloader.
"Stubby" <William.Plummer-NOSPAM-@comcast.net> wrote in message
news:GoSdncN4_o3XDyTfRVn-3w@comcast.com...
> Greg Hayes/Raxco Software wrote:
> > If a file is
> > in 1000 logical fragments, then 1000 logical requests have to be made to
the
> > hard drive controller.
> I admit I don't know how NT and following work inside, but I am certain
> that DOS did sector-at-a-time. Later versions had a queue of pending
> write-operations but nothing coalesed the pending operations.
>
> Conversely, if a file is logical contiguous, it
> > means that only 1 logical request has to be made to the hard drive
> > controller. The performance improvement comes in because you are only
> > making 1 request - not 1000 requests.
> The cost of initiating a transfer is very low, insignificant. The cost
> of actually executing a transfer is high because you need to wait for
> the arm to position to the right cyclinder and for the disk to turn to
> the right place. Of course the overhead due to physical movement is
> non-existent on a solid-state disk. This is what prompted my original
post.
>
> If you are correct, current disk controllers must be able to start a
> transfer, wait until it is almost ready to start, look at the disk queue
> to see if more contiguous sectors have been added, and update the number
> of sectors in the transfer if possible. Some how I don't think they are
> this clever, but I could easily be wrong. But this is what it would
> take to make defragmenting have an effect on efficiency.
>
> If you could post a URL for how disk controllers work these days, I sure
> would like to see it. I have some ideas for performance improvements
> that are guaranteed to work but I need to understand all the low level
> stuff as it exists today. Thanks.