RE: ASM is single point of failure ?

From: Kevin Closson <kevinc_at_polyserve.com>
Date: Mon, 8 May 2006 14:31:31 -0700
Message-ID: <5D2570CAFC98974F9B6A759D1C74BAD06FEA3D@ex2.ms.polyserve.com>

 >>>On Mon, May 8, 2006 21:32, Kevin Closson said:

>>>> ...VxVM runs in kernel mode. Kill(1) sends a signal. You
>>>What is less plesant with ASM, actually, is a kind of buffered "I/O"
>>>between database and ASM instance processes. I mean, the
>>>blocks should be shipped between them somehow, right? So
>>>some additional CPU cycles are definitely there as well as
>>>more syscalls, context switches, ipc delays.

The misconceptions are the most frightening aspect of ASM.

People are willing to use it and have ZERO idea what it is. Oddly, I find people fearing things that do not relate whatsoever (such as the above), but ignoring the bold, glaring issues. It just bogles my mind.

ASM is not in the I/O code path. Huh, what did he say? Oracle positions and performs reads and writes and that is done with either libC, LibODM or if you are a real experimental type ASMLib (on Linux). There are no comms between a shadow process and an ASM process for I/O (e.g., db file sequential read, direct path write, etc). There is for metadata ops, such as adding a datafile, or adding an extent, dropping a datafile, etc...

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

Received on Mon May 08 2006 - 16:31:31 CDT