AIX Host Specifics
From Tecniq
Contents |
fget_config
Example command outputshown below, use to identify which DS luns map to which AIX hdisk devices
# fget_config -vA ---dar0--- User array name = '181470A-130001T-CRIS_Storage' dac0 ACTIVE dac1 ACTIVE Disk DAC LUN Logical Drive utm 31 hdisk2 dac0 0 heartbeatvg hdisk3 dac1 1 oraclevg hdisk4 dac0 2 crisvg hdisk5 dac1 3 asmvg hdisk6 dac0 4 oraflashvg
AIX operating system settings
The following section outlines the settings that can affect performance on an AIX host. We look at these in relations to how they impact the two different workload types.
Transaction settings
Early AIX driver releases allowed for the changing of the cache read-ahead through the logical drive attribute settings; this has been discontinued with current releases, and is now set by the DS4000 Storage Server value, and can now only report the value from the operating system. All attribute values that are changeable can be changed using the chdev command for AIX. See the AIX man pages for details on the usage of chdev.
For the logical drive known as the hdisk in AIX, the setting is the attribute queue_depth:
chdev -l hdiskX -a queue_depth=Y -P
In the above example “X” is the hdisk number, and “Y” is the value for queue_depth you are setting it to.
For the HBA settings, the attribute num_cmd_elem for the fcs device is used. This value should not exceed 512:
chdev -l fcsX -a num_cmd_elem=256 -P
Throughput based settings
In the throughput based environment, you would want to decrease the queue depth setting to a smaller value such as 16. In a mixed application environment, you would not want to lower the “num_cmd_elem” setting, as other logical drives may need this higher value to perform. In a pure high throughput workload, this value will have no effect. AIX settings which can directly affect throughput performance with large I/O blocksize are the lg_term_dma, and max_xfer_size parameters for the fcs device.
Note that setting the max_xfer_size affects the size of a memory area used for data transfer by the adapter. With the default value of max_xfer_size=0x100000, the area is 16 MB in size, and for other allowable values of max_xfer_size, the memory area is 128 MB in size.
AIX LVM impact
AIX uses Logical Volume Manager (LVM) to manage the logical drives and physical partitions. By default, with standard and big VGs, LVM reserves the first 512 bytes of the volume for the Logical Volume Control Block. Therefore, the first data block will start at an offset of 512 bytes into the volume. Care should be taken when laying out the segment size of the logical drive to enable the best alignment. You can eliminate the Logical Volume Control Block on the LV by using a scalable VG, or by using the -T 0 option for big VGs. Additionally, in AIX, file systems are aligned on a 16K boundary. Remembering these two items helps when planning for AIX to fit well with the DS4000 segment size. JFS and JFS2 file systems intersperse inode data with the actual user data, and can potentially disrupt the full stripe write activity. To avoid this issue, you can place files with heavy sequential writes on raw logical volumes.
With AIX LVM, it is generally recommended to spread high transaction logical volumes across the multiple logical drives that you have chosen, using the maximum interpolicy setting (also known as maximum range of physical volumes) with a random ordering of PVs for each LV. Ensure that your logical drive selection is done as recommended above, and is appropriate for the RAID type selected.
In environments with very high rate, sequentially accessed structures and a large I/O size, try to make the segment size times the (N-1 for RAID 5, or N/2 for RAID 10) to be equal to the application I/O size. And keep the number of sequential I/O streams per array to be less than the number of disks in the array.
Using fast I/O failure for fibre channel devices
I/O failover occurs when a fibre channel adapter driver detects a link failure on the fabric between the switch and the DS4000 storage subsystem. You can change the failover characteristics by setting the fscsi device attribute fc_err_recov to one of the following settings.
fast_fail Enables fast I/O failure. If the fibre channel adapter driver detects that a link was lost between the switch and the DS4000, it waits a few seconds to allow the fabric to stabilize. If fast_fail is specified, the adapter then begins failing all I/O at the adapter driver if it detects that the device is not on the fabric. Any new I/O or future retries of the failed I/O are failed immediately. Fast I/O failure can be useful in multipath configurations. It can decrease the I/O fail times due to the loss of a link between the storage device and the switch, and can allow faster failover to alternate paths.
delayed_fail Default setting. If delayed_fail is specified, I/O failure proceeds as normal; retries are not immediately failed, and failover takes longer than it does if fast_fail is specified. In single-path configurations, especially configurations with a single path to a paging device, the delayed_fail setting should be used.
Example: You can enable fast I/O failure by setting this attribute, as shown in the following example. Be sure to stop all I/O and put the fscsi device into a Defined state before you set the attribute.
chdev -l fscsi0 -a fc_err_recov=fast_fail
Notes: 1. The fast_fail attribute only affects failover that occurs between the switch and the DS4000 storage subsystem. It does not affect failover that occurs between the host and the switch. 2. Set the fast_fail attribute on each HBA that is configured to the DS4000 storage subsystem. 3. You can use fast I/O failure only in a SAN environment. You cannot use it in a direct-attach environment.
Using dynamic tracking of fibre channel devices
You can dynamically track fibre channel devices, which allows the dynamic movement of a fibre channel path between the fabric switch and the DS4000 subsystem by suspending I/O for 15 seconds while the move occurs. Exception: You can only use dynamic tracking on AIX 5.2 and 5.3.
You can enable or disable dynamic tracking by setting the fscsi device attribute dyntrk to one of the following settings:
yes Enables dynamic tracking. If dynamic tracking is enabled, the fibre channel adapter detects when the fibre channel node port ID of a device changes. It reroutes the traffic that is destined for that device to the new worldwide port name (WWPN) while the devices are still online. For example, you can move a cable from one switch port to another while the devices are still online, and no failover occurs if you complete the move within 15 seconds. After 15 seconds, failover occurs. Tip: The ports must be in the same zone on the same switch. no Default setting. If dynamic tracking is not enabled, you must take the devices offline before you move a cable from one port to another. Otherwise, failover occurs. Example: You can enable dynamic tracking by setting this attribute, as shown in the following example. Be sure to stop all I/O and put the fscsi device into a Defined state before you set the attribute.
chdev -l fscsi0 -a dyntrk=yes
Notes: 1. Set the dyntrk attribute on each HBA that is configured to the DS4000 storage subsystem. 2. You can use dynamic tracking only in a SAN environment. You cannot use it in a direct-attach environment.
