OK, agreed, but can aynone give me an answer to my original question, which is "How do I tell exactly where a file resides      on a drive, is there an address location or something of that nature and if so how does it relate to the layout of the drive?"

Jack
-----Original Message-----
From:   Bill Lancaster [SMTP:[log in to unmask]@inetgw]
Sent:   Friday, April 10, 1998 12:34 PM
To:     [log in to unmask]@inetgw
Subject:        Re: File Placement

Hi John and Jack,

The issue isn't so much putting the data on the outer edges as much as it
is making sure the data is a "compacted" as it can be.  On Winchester
drives the heads do *not* withdraw but stay in place where the last access
occurred.  My example is that if you have a drive 50 percent full and it's
completely fragmented (1 page permanent space, 1 page free space, 1 page
perm, page free etc) the average head travel distance is r/2 where "r" is
the radius of the platter.  If you completely condense the drive, the
average head travel becomes less than r/4.  Why less than???  Because the
outer tracks are "denser" than the inner tracks (e.g. more sectors per
track).  (This isn't a track-by-track issue; it's a zone-by-zone issue
where there are multiple tracks per zone but each track in a zone has the
same sectors-per-track).

I have often recommended putting the more heavily accessed files on the
outer edges of the disks but there are several issues which impact whether
or not this will be a performance booster (my two favorite words are "It
Depends" - actually, I now have two other words in my lexicon of favorites:
"Business Casual" :-)

Some of these issues are:

1.  What is your data locality like?  In other words, are you accessing
data randomly across your disks?  Or are you concentrating on a small set
of files?

2.  How much free space do you have?  The less free space, the more the
heads are going to move anyway, especially when considering data locality.

3.  How well is MPE eliminating physical disk I/O?  In SOS/3000 you can
look at the "Read Hit Percentage" metric.  This is the number of read
requests satisfied in main memory.  The higher this number already is, the
less overall impact you will experience in changing how data is retrieved
from disk (on reads, anyway).

4.  If you are heavily Image-oriented, how efficient are the I/O's you're
already doing?  Keep in mind that there is no such thing as a "good" I/O.
Packing the details, having proper paths and "managing" master hashing will
be far more effective in that they contribute to the overall reduction of
disk I/O and eliminating disk I/O's is far better than making existing ones
faster.


That all being said, should you find you have an environment where placing
certain files on the outer edges of the platters is advantageous, the best
time to do this is when you reload the volume set in question.  The
disadvantage of doing this data placement activity is that you can blow it
all away with capacity changes etc...

The bottom line is that, while there may be a (slight-to-small) performance
advantage, usually the operational issues make it not very worthwhile.

Hope this helps.

Bill Lancaster


At 10:38 AM 4/10/98 -0700, John Clogg wrote:
>I'm glad this question was posted, because it brings up an issue I have
>wondered about for some time.  I have heard the recommendataion that
>files be concentrated on the outer tracks of a disc.  What is the reason
>for this?  What is special about the outer tracks?  Does it have to do
>with the assumption that the drive "parks" the heads at the outer edge
>when idle?
>
>>>> "Calandra, Jack" <[log in to unmask]> 04/10/98 07:43am >>>
>
>        Can anyone tell me how to tell exactly where on a platter a file is
>located?  I am using DBGeneral's Disk Space Manager to manage my
>platters.  I would like to make sure that my application critical KSAM files
>are close to the outer edge of the disks, but I am unsure of where the
>files are placed.   Thanks in advance.
>
>        JACK CALANDRA
>        AIG
>
>