Sun Storage 2500, 2500-M2 and 6000 Arrays May Reboot Due to Ancient I/O

Asset ID:	1-79-1663839.1
Update Date:	2016-06-30
Keywords:

Solution Type Predictive Self-Healing Sure

Solution 1663839.1 : Sun Storage 2500, 2500-M2 and 6000 Arrays May Reboot Due to Ancient I/O

Related Items


Sun Storage B220 Disk Subsystem
 Sun Storage Flexline 280 Array
 Sun Storage D178 Disk Subsystem
 Sun Storage 2540-M2 Array
 Sun Storage 2510 Array
 Sun Storage 2540 Array
 Sun Storage 6140 Array
 Sun Storage B250 Array
 Sun Storage B280 Array
 Sun Storage Flexline 210 Array
 Sun Storage 6780 Array
 Sun Storage 6180 Array
 Sun Storage 6130 Array
 Sun Storage Flexline 380 Array
 Sun Storage 2530 Array
 Sun Storage 2530-M2 Array
 Sun Storage B150 Array
 Sun Storage 6540 Array
 Sun Storage D240 (STK) Array
 Sun Storage D173 Disk Subsystem
 Sun Storage Flexline 240 Array
 Sun Storage D280 Disk Subsystem
 Sun Storage 6580 Array

Related Categories


PLA-Support>Sun Systems>DISK>Arrays>SN-DK: FLX300_65xx_6780

OEM confidential document provided as Oracle internal knowledge document for informational purposes only.  Please do not distribute.

Applies to:

Sun Storage 6540 Array - Version Not Applicable and later
Sun Storage 6140 Array - Version Not Applicable and later
Sun Storage B280 Array - Version Not Applicable and later
Sun Storage Flexline 240 Array - Version Not Applicable and later
Sun Storage Flexline 280 Array - Version Not Applicable and later
Information in this document applies to any platform.

Purpose

The intent of this document is to explain what is an "Ancient I/O" on a Sun Storage 2500, 2500-M2 and 6000 Array.

Details

A reboot due to ancient IO being aborted is basically the firmware triggering a deadlock prevention mechanism.

When the disk array controller detects that an I/O has stayed in the system for an inappropriate amount of time (usually 5 minutes), something has obviously gotten stuck, often times the IOSched task may have been blocked or there maybe two IOs waiting on each other for some reason. This situation can only be cleared by a reboot, which is what the firmware triggers to clear the deadlock.

The whole thing is triggered when an I/O request is ABORTED by an attached host system to the disk array. When this happens, the controller firmware will flag the I/O request as having been aborted. However, no attempt is made to immediately purge the request from the system. Instead, the controller firmware that handles all host interactions (e.g. for data transfers to/from the host, for completion/response handling, etc.) will always check to see if the I/O that is associated with a request for host interaction has been aborted. If so, that is the point at which more comprehensive termination and cleanup will occur. Note that, from a host standpoint, an aborted I/O will never be seen again, even though from a controller firmware standpoint, an aborted I/O may continue to make some progress. Only when the I/O request "surfaces" from the underlying RAID engine for some form of host interaction will it be fully purged from the system.

The five possible causes of an “ancient I/O” status are:

Incorrect Host Configuration.
Deadlock within the subsystem firmware related.
Channel related error (host side or drive side).
Improperly configured systems (Using up all of the buf structures because segment size of a volume is not optimal for I/O).
Slow drives.

If the disk array is not at the latest firmware level, then it is highly recommended that the array be upgraded to the latest available firmware levels in a scheduled maintenance window. If disk array is currently at the latest firmware levels, then please contact Oracle Support for more detailed troubleshooting and analysis.

Do you still have questions? You can use My Oracle Support Communities. Communities put you in touch with industry professionals like yourself. They are monitored by Oracle support engineers, so you can expect reliable and correct answers. Ask questions and see what others are asking about in the Disk Storage 2000, 3000, 6000 RAID Arrays & JBODs Community.

Attachments

This solution has no attachment