Recommendations Common to ASM and File System

Unlike traditional storage, IOPS are not a function of LUN count. In other words, you get the same IOPS capacity with 1 LUN as you do with 100. However, since it is often convenient to monitor database performance by I/O type (for example, LGWR, DBWR, TEMP), we recommend creating ASM disk groups or file systems dedicated to these individual workloads. This strategy allows you to observe the characteristics of each I/O type either at the command line with tools like iostat and pureadm, or with the Everpure GUI.

select name, allocation_unit_size/1024/1024 as "AU", state, type, round(total_mb/1024,2) as "Total", round(free_mb/1024,2) as "Free"

  from v$asm_diskgroup;

SQL > sqlplus / as sysasm
SQL > alter diskgroup DATA set attribute 'thin_provisioned'='TRUE';

SQL > alter diskgroup DATA rebalance with balance compact wait;

1. Check the ASM Filter Driver Certification Matrix for supported Linux kernel versions.
2. If ASMLib is installed and configured for an existing Oracle ASM installation, then it must be explicitly uninstalled before installing and configuring Oracle ASMFD.
3. Make sure you are on a database version that includes a fix to Bug 28378446 - Using AFD and Attribute "thin_provisioned=true" Can Cause Database Corruption.

   Method 2 - ASM Disk Group Reorganization

   As discussed above, using the ASM Filter Driver with the THIN PROVISIONED attribute set to true can reclaim space when the tablespace is dropped. However, ASM Filter Driver is relatively new and there are many Oracle database installations that either have not or, for various reasons can not make the switch to using ASM Filter Driver. Moreover, there could be cases where the database has a large amount of free space due to dropped or truncated objects but it is not practical to drop the tablespace. For such scenarios, the method described below can be used.

   Add a New Disk and Drop the Old One

   Identify the disks that have a large amount of free space that can be reclaimed, and the disk group they are part of.

   Execute the following command to replace the existing bloated disk in the disk group with a new disk. This SQL also initiates a ASM rebalance of the blocks from the old disk to the new disk. As the wait clause is used, control will be returned only after the operation is complete.

   
   SQL > alter diskgroup DATA add disk '/dev/oracleasm/disks/DATA2' name DATA2
         drop disk DATA1
         rebalance power 16 wait;
   

   So at the end of this command, the old disk will no longer be part of the disk group and can be safely deleted on the FlashArray.

   Note: Oracle recommends that all the disks in a disk group be of the same size. Starting with version 12.2, it enforces this condition for disk groups created with Normal or High redundancy.

   ASMLib and Alternative

   ASMLib is an Oracle-provided utility that allows you to configure block devices for use with ASM. Specifically, it marks devices as ASM disks and sets their permissions so that the o/s account that runs ASM (typically either grid or oracle) can manipulate these devices. For example, to create an ASM disk named MYASMDISK backed by /dev/dm-2 you would issue the command:

   
   /etc/init.d/oracleasm createdisk MYASMDISK /dev/dm-2

   Afterward, /dev/dm-2 appears to still have the same ownership and permissions, but ASMLib creates a file /dev/oracleasm/disks/MYASMDISK owned by the O/S user and the group is identified in /etc/sysconfig/oracleasm. Tell the ASM instance to look for potential disks in this directory through the asm_diskstring initialization parameter.

   In case you do not want to use ASMLib, you can certainly use the UDEV rules, which is the basis for ASMLib.

   Dedicated ASM Disk Froups for Each Database

   ASM allows files from multiple databases to be placed under a common disk group. Create dedicated ASM disk groups for each database.

   ASM SPFILE Location

   When a new database is created using the Database Configuration Assistant (dbca), ASM password (orapwasm) and spfile are typically placed in the +DATA diskgroup. It is recommended that database files not be placed on the same disk group as ASM files because restoring a database from a snapshot backup will overwrite the ASM configuration files as well. This will have the unintended consequence of crashing and/or corrupting the ASM instance.

   Disable ASM Compact

   ALTER DISKGROUP DATA_PRODDB SET ATTRIBUTE '_rebalance_compact'='FALSE';

   File System Recommendations

   There is no significant performance penalty for using a file system instead of ASM. As with ASM, we recommend placing data, redo, and the fast recovery area (FRA) onto separate volumes to ease administration. We also recommend using the ext4 file system and mount it with discard and noatime options. Below is a sample /etc/fstab file showing mount points /u01 (for oracle binaries, trace files, etc.), /oradata (for datafiles) and /oraredo (for online redo logs).

   

   The main page for mount describes the /discard flag as follows:

   discard/nodiscard

   Controls whether ext4 should issue discard/TRIM commands to the underlying block device when blocks are freed. This is useful for SSD devices and sparse/thinly-provisioned LUNs, but it is off by default until sufficient testing has been done.

   Mounting the ext4 file system with the discard flag causes freed space to be trimmed immediately, just as the ASRU utility trims the storage behind ASM disk groups.