HP XP RAID Manager Software - Manual (Page 144)

Table 33 Relationship between Continuous Access Synchronous pair statuses and fence levels (continued)
Fence level and write response Continuous Access Synchronous pair status of
volume
Async
4
Never
3
Status
2
Data
1
and are also noted in a bitmap for future transfer. Due to the use of a bitmap in the suspend state, data consistency is not
assured during a copy state resync.
•
PAIR: If the P-VOL and S-VOL are both in a PAIR state, mirror consistency is not assured (may be behind) but data consistency
is assured (what has reached the S-VOL is in the proper order).
Setting the fence level
Data fence level
Figure 57 (page 144) shows the relationship between redo log files (journal) and data files. If the
S-VOL takes over from the P-VOL in the status shown in Figure 57 (page 144) (where two errors
have occurred), the secondary host leaves data (V) unprocessed in the roll-back processing and
cannot be recovered completely. Therefore, the fence level of a redo log file must be defined as
data. Once the fence level is set to data, the P-VOL returns an error if data may possibly be
inconsistent when a write request is issued by the host. Since writing to the data file is not executed
due to a write error of the redo log file, the log file stays consistent with the data file. However,
when the fence level is set to data, a write I/O error occurs even in the case where operation is
suspended due to an error in the S-VOL. Accordingly, duplication becomes meaningless when the
S-VOL takes over. Thus, applications using paired volumes with the data fence level should be able
to handle write I/O errors properly. For example, the Oracle application creates multiple redo log
files by itself (three by default). The fence level can be set to data in this case in which disk errors
are permissible by creating multiple files.
Figure 57 Relationship between logs (journal) and data in paired status
Never fence level
Because most UNIX file systems (excluding JFS and VxFS) have no journal files, the fence level
should be defined as Never. When a takeover by the S-VOL occurs, fsck is executed on the volume
and the file system is cleaned up, even if the S-VOL is undefined at the secondary host. The data
that is lost depends on how much differential data is contained in the P-VOL when the S-VOL is
suspended. During operation, error recovery should be performed when the suspended status
(PSUE or PDUB) is detected (when one error occurs).
Snapshot operations
Snapshot normally creates virtual volumes for copying on write without specifying LUNs as S-VOLs.
However, to use a Fast Snap or Snapshot volume via the host, it is necessary to map the Snapshot
S-VOL to a LUN. Therefore, RAID Manager provides a combined command to enable the user or
application to use the same RAID Manager command in order to maintain Business Copy
compatibility.
Snapshot uses two techniques, one called “V-VOL mapping” (or virtual volume mapping) and the
other is “Snapshot using copy on write” or “Copy-on-Write Snapshot.” Snapshot volumes are also
put into pooling volumes called a “Snapshot pool,” and a Fast Snap or Snapshot pool is specified
as a pool ID when a Fast Snap or Snapshot is made. Snapshot and volume mapping is illustrated
in
Figure 58 (page 145) .
144 Data replication operations with RAID Manager