Re: ASYNCH I/O

Doug, in standard UNIX there is no such thing as an Asynch I/O. The process always waits for the I/O to be completed before control is returned to the process by the OS. This is regardless of the OS actually using a cache or not to speed up this process.

Asynch I/O is an option added on to some UNIX versions to allow a process to hook
a jump for when the OS has finished the I/O. The process can continue executing and the jump is taken when the OS has finished the I/O, be it cached or not.

Note that I'm separating the OS use of a cache from the OS reaction to a I/O request. Using a cache is internal to the OS and should be/is transparent to the user processes, unless the user process has elected to know about it.

Now, of course there are back doors...

In some versions of UNIX, Asynchronous I/O is a separate option that needs to be configured. Of course, ORACLE doesn't know if this has been configured or not. In such cases, you have the option of telling ORACLE to use the Asynch I/O system calls instead of the standard I/O, so that it can take advantage of the
thing if it's there.

In a nutshell, that's what it all means. There are slight detail variations here and there,
depending on the OS, file system and the particular versions thereof.

Some OS versions will do the Asynch I/O using its own background processes. Others will not. It all depends.

Bottom line is: if you are fairly sure that the OS you're using has Asynch I/O activated AND configured, then you turn on Asynch I/O in ORACLE and forget the multiple db-writers. But you GOTTA be fairly sure that all of that is true, otherwise
the OS will channel Asynch I/O calls from ORACLE into standard I/O and not tell you
about it, with obvious consequences to performance...

HTH --
Cheers
Nuno Souto
nsouto_at_nsw.bigpond.net.au.nospam
Is there a nospam domain?
http://www.users.bigpond.net.au/the_Den Doug Cowles <dcowles_at_bigfoot.com> wrote in message news:377FA9F6.5844956E_at_bigfoot.com...
> Ran across a statment that multiple database writters are useless if you
> are using
> Asynch I/O. Does this statement apply to OS Asynch I/O, or the Oracle
> simulated
> Asynch I/O created with USE_ASYNCH_IO = true, or is it true at all?
>etc.
Received on Mon Jul 05 1999 - 06:58:34 CDT