Re: Linux, i/o scheduler and ASM

  • From: Kevin Closson <ora_kclosson@xxxxxxxxx>
  • To: "oracle-l@xxxxxxxxxxxxx" <oracle-l@xxxxxxxxxxxxx>
  • Date: Fri, 22 Apr 2011 14:21:07 -0700 (PDT)

The linux I/O scheduler has little to do with random small I/Os (e.g., db file 
sequential read, db file parallel write,etc). They have most to do with 
opportunities to coalesce adjacent I/O requests thus condensing the number of 
physical transfers with larger payloads. I wouldn't ever expect to see much. We 
had reasons to explore this in cellsrv (the Exadata Storage Server software 
executable). However, if your large I/Os are buffered in hugepages memory there 
is even less the I/O schedulers can do to help/harm your I/O.





________________________________
From: "CRISLER, JON A (ATTCORP)" <JC1706@xxxxxxx>
To: "oracle-l@xxxxxxxxxxxxx" <oracle-l@xxxxxxxxxxxxx>
Sent: Mon, April 18, 2011 8:51:01 AM
Subject: Linux, i/o scheduler and ASM

 
Has anybody played around with changing the Linux I/O scheduler when using ASM 
(11gR2) ?  I am wondering if any performance differences are to be found in 
using  CFQ vs. Deadline vs other options, especially in a large RAC / SAN 
environment.  Right now we are using CFQ, and in my previous experiments on 
filesystems I did not get any noticeable difference in changing to deadline.

Other related posts: