IBM System Storage DS4000, Hardware Manual

Page: 256 / 566

238
IBM Midrange System Storage Hardware Guide
4.10.5 Performing advanced functions on logical drives (LUNs)
This section details the advanced functions that can be performed on logical drives.
Pre-read redundancy check
RAID arrays are created to provide tolerance against disk failure and errors. To allow recovery
from failures, the RAID array stores additional data called redundancy data. This redundancy
data can potentially become inconsistent from time to time. Redundancy data is “consistent” if
every portion of the redundancy group can be reliably and consistently reconstructed in the
event of a failure to the array, and “inconsistent” if it cannot be read or the consistency cannot
be verified.
With Storage Manager, a pre-read redundancy check feature is available. The pre-read
redundancy feature allows the consistency of redundancy data of a RAID array to be
optionally checked prior to returning data for host read requests. The consistency check
cannot determine whether the data itself is correct. It does, however, determine that a
redundancy consistency error has occurred within the extent of the read command. Once an
inconsistency is discovered, an error is reported in the storage subsystem, and no read data
is returned, depending on the data inconsistency error.
This feature can be activated on a logical drive basis, which is created as a RAID array that
supports redundancy information. The feature allows data verification in environments where
data consistency is a key requirement. This feature cannot be enabled for logical drives that
do not have any redundancy data.
When a read request is issued to the storage subsystem, the controller verifies that the
redundancy group is consistent for the extent of the data specified in the read request. If the
controller found that the redundancy group data is in a consistent state, the read request is
returned to the host successfully. If the redundancy group data is found to be in an
inconsistent state, the read request is returned to the host with a check condition status. The
consistency check is only performed for data not already in cache.
In order to verify the consistency of the redundancy group, the entire redundancy group must
be read from disk into cache. Should no inconsistency be found, all user data that comprises
the redundancy group is left in cache in order to potentially satisfy subsequent cache hits.
The data left in cache might include data that is outside the extent of the original read request.
Such data does not require a consistency check, because the data is already in cache.
During the process of verifying the consistency of the redundancy group, a media error or
unreadable sector might be encountered within the extent of the redundancy group. When
such an error occurs, the RAID controller attempts to perform a reconstruction of the data,
using the redundancy information of the data. If this operation is successful, the consistency
check indicates that the redundancy group is consistent and the read data is returned to the
host.
During the process of verifying the consistency of the redundancy group, an unrecoverable
read error might be encountered. When such an error occurs, the consistency check fails.
However, the original read request continues to be processed. If the entire extent of the
request is still readable, the read data is returned to the host. If some portion of the original
read request is unavailable, the read command fails.
Note: Should a drive associated with an array become degraded, the pre-check feature is
disabled.