Page 1 of 3

"> How to use Concurrent I/O on HP-UX and improve throughput on an Oracle single-instance database [ID 1231869.1] Modified 08-OCT-2010 In this Document Goal Solution What is Concurrent I/O ? How to enable Concurrent I/O ? Concurrent I/O requirements Using Concurrent I/O with Oracle Summary References Type HOWTO Status PUBLISHED

Applies to:
Oracle Server - Enterprise Edition - Version: 10.2.0.1 to 11.2.0.2 - Release: 10.2 to 11.2 HP-UX PA-RISC (64-bit) HP-UX Itanium HP-UX PA-RISC - HP-UX 11iv3 September 2009 Operating Environment Update Release HP-UX Integrity Blade Server - HP-UX 11iv3 September 2009 Operating Environment Update Release VxFS 5.0.1 - Using OnlineJFS 5.0.1

Goal
How to use Concurrent I/O on HP-UX and improve throughput on an Oracle single-instance database.

Solution
What is Concurrent I/O ?
Concurrent I/O allows multiple processes to read from or write to the same file without blocking other read(2) or write(2) calls. POSIX semantics requires read and write calls to be serialized on a file with other read and write calls. With POSIX semantics, a read call either reads the data before or after the write call occurred. With Concurrent I/O, the read and write operations are not serialized as in the case of a character device. This advisory is generally used by applications that require high performance for accessing data and do not perform overlapping writes to the same file. It is the responsibility of the application or the running threads to coordinate the write activities to the same file when using Concurrent I/O.

How to enable Concurrent I/O ?

Concurrent I/O can be enabled in the following ways: A. By using the -o cio mount option. The read(2) and write(2) operations occurring on all of the files in this particular filesystem will use Concurrent I/O. - Steps for new filesystems created using 5.0.1 OnlineJFS: # mount -F vxfs -o cio <device_special_file> <mount_point> - Steps for already existing filesystems which were mounted without Concurrent I/O or created with older vxfs versions: Existing filesystems (older filesystems created or filesystems that were not mounted with Concurrent I/O option) will have to be unmounted and mounted again with "-o cio" to enable Concurrent I/O. Note that remount command/option should not be used while mounting a filesystem with "-o cio". 1. Unmount the filesystem # umount <mount_point> 2. Upgrade to VxFS 5.0.1 with 5.0.1 OnlineJFS installed on the system Refer to Veritas 5.0.1 Installation Guide on http://docs.hp.com for detailed upgrade instructions. 3. Mount the filesystem with -o cio option # mount -F vxfs -o cio,<other_options_as_needed> <device_special_file> <mount_point> Concurrent I/O is a licensed feature of VxFS. If -o cio is specified, but the feature is not licensed, the mount command prints an error message and terminates the operation without mounting the filesystem. NOTE: Do not use "-o cio" and "-o mincache=direct,convosync=direct" together. Use either Direct I/O or Concurrent I/O. Using Direct I/O and Concurrent I/O("-o mincache=direct,convosync=direct,cio") may cause performance regression. B. By specifying the VX_CONCURRENT advisory flag for the file descriptor in the VX_SETCACHE ioctl command.

https://support.oracle.com/CSP/main/article?cmd=show&type=NOT&doctype=HOWTO... 29/01/2011

Page 2 of 3

Only the read(2) and write(2) calls occurring through this file descriptor use concurrent I/O. The read and write operations occurring through other file descriptors for the same file will still follow the POSIX semantics. Concurrent I/O (CIO) can be set through the file descriptor and ioctl() operation using the VX_SETCACHE ioctl command with the VX_CONCURRENT advisory flag. The VX_CONCURRENT advisory can be set via the VX_SETCACHE ioctl descriptor on a file. For example: ioctl(fd,VX_SETCACHE,VX_CONCURRENT); where fd is the file descriptor.

Concurrent I/O requirements

With Concurrent I/O, the read and write operations are not serialized. It is the responsibility of the application or the running threads to coordinate the write activities and ensure they are to non-overlapping blocks of the same file. To gain maximum throughput, Application must perform non-overlapping writes to the same file. Performance increases if application write offsets are block aligned and size of I/Os are in multiple of device block size. Concurrent I/O bypasses inode locking and hence application (or database used) must have its own inode-locking (serialization) mechanism for multiple writers. The starting file offset must be aligned to a 1024-byte boundary. The ending file offset must be aligned to a 1024-byte boundary, or the length must be a multiple of 1024 bytes. NOTE: If the Concurrent I/O alignment requirements are not met properly, then I/Os defaults to data synchronous I/O which could cause performance degradation. If an application issues overlapping writes(on a filesystem with Concurrent I/O enabled) to the same file without having its own serialization mechanism, behavior of write(2) is undefined and may lead to data loss.

Using Concurrent I/O with Oracle

It is recommended to place the Oracle binaries ($ORACLE_BASE directory) on a separate filesystem mounted with default options. Placing the Oracle Database's datafiles on filesystems mounted with Concurrent I/O ("-o cio) delivers performance very close to that of raw logical volumes. Placing Oracle binaries ($ORACLE_BASE directory) on a filesystem mounted with "cio" may cause data loss and other unexpected problems. Hence mounting Oracle binaries ($ORACLE_BASE directory) on a filesystem mounted with "cio"option is not supported. Concurrent I/O can significantly improve performance of a filesystem based Single instance Oracle database installation. Concurrent I/O performs at between 93-99% of raw logical volumes. Thus, Concurrent I/O provides performance very close to that of raw logical volumes in addition to the manageability benefits provided by a filesystem. Concurrent IO is not expected to provide a performance benefit over direct IO when used with online and archived redo logs.

Summary
Option 1. Production and other performance critical databases: 3 VGs and 5 filesystems.
VG 1 vgSIDlog Filesystem Contents origlogA Online redo logs group A Online redo logs group B Database files Archived redo logs Database binaries Concurrent or Direct IO ? Direct Filesystem Block Size Default Notes Dedicated VG can (i) improve performance as writes dont have to wait on shared SCSI queue as would happen if db files and logs were in same VG (ii) simplify I/O performance analysis (log writer response times).

origlogB 2 vgSIDdata Data1 3 vgSIDfs Archfile Home

Direct Concurrent Direct Default

Default 8KB Default Default Binaries could be placed on a separate VG however there is no performance benefit in doing so Dedicated VG allows BCs

Option 2. Non-production and other less performance sensitive databases; 2 VGs and 5 filesystems.
VG Filesystem Contents Online redo logs group A Online redo logs group B Database files Archived redo logs Database binaries Concurrent or Direct IO ? Direct Direct Concurrent Direct Default Filesystem Block Size Default Default 8KB Default Default Binaries could be placed on a separate VG however there is no performance benefit in doing so If BC is used db files should be placed into a dedicated VG i.e. as per production layout. Notes

1 vgSIDdata origlogA origlogB Data1 2 vgSIDfs Archfile Home

https://support.oracle.com/CSP/main/article?cmd=show&type=NOT&doctype=HOWTO... 29/01/2011

Page 3 of 3

References

Related Products

Oracle Database Products > Oracle Database > Oracle Database > Oracle Server - Enterprise Edition

Keywords THROUGHPUT; HP-UX; I O

Back to top

https://support.oracle.com/CSP/main/article?cmd=show&type=NOT&doctype=HOWTO... 29/01/2011