IBM DS6000 - Manual (Page 361)

Chapter 25. Global Mirror performance and scalability 337
There is potential impact on the Global Copy data replication operation, depending on
whether persistent memory or non-volatile cache is over-committed in the secondary storage
disk subsystem. In this situation, the FlashCopy source tracks might have to be preserved
first to the FlashCopy target volume, before the Global Copy write completes. Usually,
however, all writes are quick writes to cache and persistent memory.
A write I/O to the FlashCopy source volume also triggers the maintenance of a bitmap for the
source volume, which is created when the FlashCopy volume pair is established with the
start
change recording
attribute. The maintenance process means that you only have to replicate
the changed recording bitmap to the corresponding bitmap for the target volume in the course
of formation of a Consistency Group. See 22.4, “Consistency Groups” on page 260, for more
information about this topic.
Note that this all applies only to write I/Os to the Global Mirror primary volumes.
25.2 Performance considerations at coordination time
When you look at the three phases that Global Mirror goes through to create a set of data
consistent volumes at the secondary site, the first question that comes to mind is whether the
coordination window imposes an impact to the application write I/O; see Figure 25-2.
Figure 25-2 Coordination time - how it impacts application write I/Os
The
coordination time
, which you can limit by specifying a number of milliseconds, is the
maximum impact to the write I/Os of an application, that you will allow when forming a
Consistency Group. The intention is to keep this time window as small as possible. The
default 50 ms might be a bit high in a transaction processing environment. A valid number
could also be a very low number in the single digit range. The required communication
between the master storage disk subsystem and subordinate storage disk subsystems is
inband, over the Global Mirror session paths between the master and subordinates. This
communication is highly optimized and allows you to minimize the potential application write
I/O impact to 3 ms, for example. This communication is performed over FCP links. At least
one FCP link is required between a Master storage disk subsystem and any potential
Subordinate storage disk subsystems. For redundancy, we suggest using two FCP links.
Maximum
coordination
time
drain
time
Maximum
2
Serialize all
Global Copy
primary volumes
Drain data from local to remote site
Perform
FlashCopy
1 3
Primary
A2
Primary
A1
B2
Secondary
B1
Secondary
C2
Tertiary
C1
Tertiary
H
o
ld
w
rit
e
I/
O
I/O
Write
Global Copy paths
Remote site Local site
Global Mirror
session paths
Global Copy paths