AIX Down Under

An IBM Oracle Technical Brief came out in May entitled Oracle Architecture and Tuning on AIX. There's a lot in it, but the good news is that if you're working on an AIX system that was a fresh install of AIX 6.1 or 7.1, then on the whole the default tuning parameters are what you need. If you're still on AIX 5.3, or if your system was migrated from AIX 5.3 to 6.1 to 7.1, you may need to revisit the tuning parameters. Do it anyway, if you can.

There are two items which caught my eye especially from this Oracle Technical Brief: Logical Volume striping, and mounting file systems using Concurrent I/O (cio).

Here's what they say:

Logical Volume Striping

AIX LVM supports two striping techniques:

• PP (Physical Partition) Spreading
• LVM Striping

PP Spreading is initiated during the creation of the logical volume itself. The result is a logical volume that is spread across multiple devices (hdisk, vpath, hdiskpower ... ) in the Volume Group. To create a LV with PPs spread equally over a set of hdisks in a VG use the following AIX command:

mklv –e x –y <lvname> <vgname>

There is at least one clear advantage of PP spreading over LVM striping. If the LV created using the PP spreading technique has to be extended, any number of disk drives (hdisks) can be added. Once the VG is extended, optimal PP re-distribution of the new disk layout is achieved using the 'reorgvg' command and can be done online.

My colleague and compatriot, Chris Gibson, recently wrote about his experiences with reorgvg, and how he diagnosed a problem he had due to lack of spare PPs in the volume group.

Back to the Oracle Technical Brief:

Extending the LV that was created using the LVM striping is a more complex process potentially requiring application downtime. The number of drives that have to be added to the VG has to be equal to the stripe width that the LV was originally defined and created with. The number of drives added is a multiple of the existing number used for the LV.

That last point is very important. It's frustrating to tell someone with two spare disks that they can't extend their three-disk stripe without an outage or buying even more disk. The first method of extending the Physical Partitions (PPs) across several hdisks is a lot easier. Or you could have a volume group on one great big LUN, but you'll need to remember to change the queue depth on the hdisk.

Concurrent I/O

Onto the section about using CIO. This is an option used when you mount the file system. There's an important point about creating file systems for redo logs.

Concurrent I/O should only be used for Oracle .dbf files (data & index, rbs or undo, system and temp), online redo logs and/or control files. When used for online redo logs or control files, these files should be isolated in their own JFS2 file system(s) that have been created with agblksize=512. File systems containing .dbf files should be created with agblksize=2048 if DB BLOCK SIZE=2k, or agblksize=4096 if DB BLOCK SIZE >= 4k or larger. Failure to implement these agblksize guidelines is likely to result in a severe performance penalty. Do not, under any circumstances, use CIO mount option for the file systems containing the Oracle binaries (!!!). Additionally, do not use DIO/CIO options for file systems containing archive logs or any other files not already discussed.

The Technical "Brief" is not very brief at all. It's quite comprehensive and unravels some of those Oracle parameters which can bamboozle non-DBAs. Once again, the brief is caled Oracle Architecture and Tuning on AIX.