Re: Async and Direct I/O On Solaris using VXFS

  • From: Grzegorz Goryszewski <grzegorzof@xxxxxxxxxx>
  • To: Posting <oracle-l@xxxxxxxxxxxxx>
  • Date: Tue, 28 Jun 2011 17:21:56 +0200

On 2011-06-28 17:02, rajesh koppada wrote:
> I was following the below link and tracing the systems calls ,
> We are using oracle version on VXFS file systems on Sun solaris
> 4294/1:         open("/users/oracle/undofile/undotbs01.dbf", O_RDWR|O_DSYNC) 
> = 10
> 4294/1:         ioctl(257, (('V'<<24)|('X'<<16)|('F'<<8)|1), 0x00000004) = 0
> any idea how to intepert this ioctl arguments ?? Just to ensure we are 
> getting Direct I/O on VXFS file systems.
Looks like a bug, do You have patch *10336129 installed?* :


Poor Oracle IO performance on VXFS for RDBMS 11gR2 compared to 10gR2 on

This problem is introduced in RDBMS on SolarisRegardless of
what you Oracle "filesystemio_options = asynch | none" is set, Oracle
issues "ioctl(fd, VX_SETCACHE, VX_DIRECT).".The ioctl effectively
disables VXFS caching, and renders considerable IO slow down compared to
10.2.x for IO events like "db file sequential read" and "db file
parallel read".


The ioctl can be observed with truss:

|21770/1: 17.5346 0.0059 ioctl(259, (('V'<<24)|('X'<<16)|('F'<<8)|1),
0x00000004) = 0|


This problem only impacts regular buffered IO on VXFS.

VXFS filesystems using ODM or mount option "mincache=direct" is not

RAC Clustered filesystems are not impacted because ODM is a requirement.


Apply Oracle patch# 10336129 to RDBMS



Przyjemna praca, wysoka placa!
Szukaj >>

Other related posts: