IBM DS6000 Manual Do Utilizador

Chapter 25. Global Mirror performance and scalability 

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 

 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.

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.

The 

, 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

Remote site

Local site