Re: Raw Devices: Increased Performance?

Jef Kennedy (jkennedy_at_oracle.com) wrote:
: In article <EiAXoAAcT$8xEwOu_at_smooth1.demon.co.uk>, David Williams
: <djw_at_smooth1.demon.co.uk> wrote:
 

: > Generally disk drive controllers themselves which have the elevator
: > seek algorithm built-in to them, also the database engine has this
: > built-in as well. Why waste time doing it an extra time ?
 

: Good point. If I remember right, the Oracle kernel has an elevator
: algorithm built in, and it was useful when disk controllers weren't doing
: the job. Nowadays, it doesn't make a lot of sense for the kernel to sort
: the blocks going to individual drives since the disk controller can do a
: better job of it. The controller knows where the head really is and sees
: all the traffic going to the drive, not just dbwr's writes. I believe
: most Oracle ports now have that Oracle elevator algorithm turned off.

[ Following based on this thread already being cross-posted: comp.sys.sequent ]

The thing I've not heard anywhere in this entire discussion is mention of "Direct I/O", which has been available under Dynix/ptx since v. 1.x (even though it was not in the on-line man pages til ptx 4.x). Direct I/O, or DIO, allows one to use a file on a filesystem like you would a raw device; that is, DIO supports "IO directly from a processes [sic] address space to a file, bypassing any buffering by the DYNIX/ptx kernel" (quoting from the DIO(2SEQ) man page on ptx 4.2.0).

We were told (by consultants from Oracle Corp., as I remember) as early as 1993 that if the DB engine utilized DIO to it's data files, that there was no more than 5% performance gain in switching to to raw data files. This allows you to get the best of both worlds: performance gain via bypassing the *NIX buffer cache, but the ease of admin. of data files on mounted filesystems.

Has this changed at all in the last 3 years? Received on Mon Jul 29 1996 - 00:00:00 CEST