RE: High disk capacity dangers


Even if this theory was correct for regular files, it would have been invalid
for Oracle DataFiles which are *precreated* (ie all disk blocks are preallocated
when the datafile is created). Of course, as you [auto]extend a datafile
or add a new datafile, you would think that the file is "fragmented".

With modern storages, striping-and-mirroring, very large disk groups,
large LUNs etc,  all of that is irrelevant.  On some storage architectures
you have no way of knowing *which* disk your data is on,  leave alone
whether the data is on the inner tracks or the outer tracks.
Any table's 1MB extent is most likely striped across multiple disks as well.

The 90% rule would be irrelevant only when filesystem management
overheads, if any, kick-in (eg when you have lots and lots [10s of thousands]
of small files {not Oracle datafiles} and you find that backing up that filesystem
itself takes a long time !).

Hemant
Hemant
At 09:35 PM Tuesday, Jesse, Rich wrote:

It was a few years ago and therefore subject to updating, but I've
typically seen the "80%" Rule.  The idea being that a physical drive
fills from the outer tracks, and the speed of the data bits as they fly
by the heads slows as the inner tracks are approached (think of standing
on the outer edge of a merry-go-round versus being in the middle of it).
Somehow, the generic "80%" was settled on as a filling point past which
the access speed of data on the inner tracks degrades "too much".



Hemant K Chitale
http://web.singnet.com.sg/~hkchital


--
http://www.freelists.org/webpage/oracle-l


Other related posts: