IBM TotalStorage DS6000 Series - Manual (Page 130)

106 DS6000 Series: Concepts and Architecture
In order for the data to be consistent, the deposit of the paycheck must be applied
before
the
withdrawal of cash for each of the checking accounts. However, it does not matter whether
the deposit to checking account A or checking account B occurred first, as long as the
associated withdrawals are in the correct order. So for example, the data copy would be
consistent if the following sequence occurred at the copy. In other words, the order of updates
is not the same as it was for the source data, but the order of
dependent
writes is still
preserved.
1. Deposit paycheck in checking account B
2. Deposit paycheck in checking account A
3. Withdraw cash from checking account B
4. WIthdraw cash from checking account A
How does Consistency Group keep data consistency?
Consistency Group operations cause the storage units to hold I/O activity to a volume for a
time period by putting the source volume into an extended long busy state. This operation can
be done across multiple LUNs or volumes, and even across multiple storage units.
In the storage subsystem itself, each command is managed with each logical subsystem
(LSS). This means that there are slight time lags until each volume in the different LSS is
changed to an
extended long busy
state. Some people are concerned that the time lag causes
you to lose data consistency, but, it is not true. We explain how to keep data consistency in
the Consistency Group environments in the following section.
See Figure 6-12. In this case, three write operations (1st, 2nd, and 3rd) are dependent writes.
This means that these operations must be completed sequentially.
.
Figure 6-12 Consistency Group: Example 1
1st
2nd
3rd
dependency
for each write
operation
Servers
Wait
Wait
Wait
LSS11
LSS12
LSS13
These write operations
are not completed
because 1
st write
operation is not
completed.
This write operation is
not completed because
of extended long busy
condition.