Improving the performance of single instance Oracle on file systems, January 2008

5
Related documents
The following documents were valuable when conducting the benchmarks and writing this white
paper.
• “Application use cases for the HP Serviceguard Storage Management Suite,” in HP Serviceguard
Storage Management Suite at http://docs.hp.com
Æ High Availability Æ HP Serviceguard Storage
Management Suite.
• “Oracle Disk Manager,” April 2001, at
http://www.oracle.com/technology/deploy/availability/pdf/odm_wp.pdf
.
• “Oracle Disk Manager” by Nitin Vengurlekar, at
http://www.oracle.com/technology/deploy/availability/pdf/nitin_ODM.pdf
• VERITAS Storage Foundation 4.1 Oracle Administrator’s Guide, at http://docs.hp.com
Æ 11i v2
(under “Operating Environments”) Æ Storage Foundation.
• VERITAS File System 4.1 Administrator’s Guide, at http://docs.hp.com
Æ 11i v2 (under
“Operating Environments”) Æ VxFS.
Oracle Disk Manager (ODM)
ODM was jointly developed by Oracle and Veritas (now Symantec Corporation) to address
performance bottlenecks when running an Oracle database on file systems. An implementation of the
ODM library was first introduced with Veritas DB Edition 3.0 for Oracle 9i in 2001.
With raw volumes or devices, HP-UX has the asynchronous device disk driver allowing kernelized
asynchronous I/O to exploit parallelism at the device level. User processes, such as Oracle log writer
(lgwr) and database writer (dbwr), do not block while waiting for I/Os to complete and the
asynchronous disk driver can intelligently sort I/O requests to streamline their execution.
On file systems, the HP-UX kernel requires the entire file to be locked to allow a write to proceed. This
file-locking serializes these accesses. Other I/O requests are thus blocked and must wait for an earlier
I/O to complete before being allowed to execute.
It is possible to partition data, dividing it into multiple files, to reduce the effect of write locking. As the
degree of partitioning increases, parallelism is improved, but additional resources are used and
managing the partitioning can become onerous.
ODM provides file system-based asynchronous I/O through a single interface mechanism
implemented in odm_io(). The library allows Oracle files to be opened and identified as ODM files.
Instead of file descriptors, ODM identifiers are pooled, reducing the kernel resources that are used
because they are shareable across Oracle processes. (A fuller description of ODM can be found in
the references cited in the Related Documents section.)
Before the release of ODM, Veritas’ Quick I/O™ provided similar functionality for Oracle 7 and 8,
but with a higher administrative overhead. By default, when mounting a Veritas file system, the –o
qio option is present. This allows special files identified with the file suffix ::cdev:vxfs: to gain
the benefit of asynchronous I/O. To facilitate this naming convention, soft links are usually employed
to point to the Oracle data files.
ODM supersedes Veritas Quick I/O and works with Oracle Database 9i and later, eliminating the
need to administer soft links because no special naming convention is required.
However, Quick I/O must not be disabled in the ODM environment because the same file device
driver (Veritas’ File Device Driver) is needed to enable the ODM functionality. That is, for ODM it is
necessary to continue to mount the Oracle file system with the VxFS mount option –o qio.