RE: Linux 2.6 I/O Scheduler & ASM/Raw

From: Kevin Closson <kevinc_at_polyserve.com>
Date: Thu, 20 Oct 2005 20:14:31 -0700
Message-ID: <B9782AD410794F4687F2B5B4A6FF3501021D4E37@ex1.ms.polyserve.com>

how do you suppose a linux process is going to perform I/O without a system call? ASM I/O is either simply libC, LibODM or ASMlib on Linux (which is still implemented in Kernel mode).  

yes yes, I know about that goofy DAFS stuff that never saw the light of day.

---

        From: oracle-l-bounce_at_freelists.org
[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Murching, Bob

	Sent: Thursday, October 20, 2005 5:46 PM
	To: 'Oracle Mailing List'
	Subject: Linux 2.6 I/O Scheduler & ASM/Raw
	
	

	Apologies in advance for the simple question but if a database

uses raw or ASM storage, is it bypassing the operating system's I/O layer, or not? For example, should the choice of scheduler (cfq, as, noop, deadline) in the Linux 2.6 affect performance of an ASM or RAW filesystem, or should it not? Likewise, are variables such as the maximum I/O size at an OS level (e.g. 512KB or 1M) and the OS block size (8KB, 16KB, ...) still meaningful with raw filesystems?

        I suspect the answer to both questions is "yes" but I'm not sure and haven't seen this spelled out clearly....

        Bob

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

Received on Thu Oct 20 2005 - 22:17:18 CDT