Extended long busy conditions
Peer-to-peer remote copy (PPRC) can be configured with the consistency group option on an LSS level or PPRC path level. When the consistency group option is enabled for the PPRC primary volume and the volume suspends unexpectedly or via a freeze operation, the volume enters an extended long busy condition for count key data (CKD) volumes or SCSI queue full condition for fixed block (FB) volumes. A freeze operation removes all paths between the frozen LSS pair and causes a subsequent suspend of all affected primary volumes.
While a primary PPRC volume is in the extended long busy or SCSI queue full condition, it does not allow further write I/O. When this event occurs, there is a window that allows external automation to freeze all dependent LSS pairs and create a crash consistent data point at the secondary site before the extended long busy condition is released again. The extended long busy condition can be released immediately after all dependent freeze operations are processed, or it will be removed automatically when the DS8000® extended long busy timeout is expired. The default extended long busy timeout is 120 seconds on CKD LCUs and 60 seconds on FB LSSs.
The freeze and unfreeze mechanism with the extended long busy condition is required to create consistent Metro Mirror secondary volumes upon (partial) replication failures. The mechanism is available with GDPS® and Copy Services Manager.
Managing the extended long busy condition
While the extended long busy condition is required to split Metro Mirror pairs consistently, it is not required for Global Mirror pairs that use a different consistency group creation mechanism. For Global Copy pairs, the extended long busy condition is not useful since these pairs are not consistent.
The lowest granularity to control the PPRC consistency group option (and extended long busy condition) is on the LSS pair level by specifying the option on the PPRC paths between the LSS pair. It is good practice to avoid sharing an LSS pair between different replication types like Metro Mirror and Global Copy to prevent the required consistency group option for Metro Mirror from unnecessarily impacting Global Copy I/O when an unexpected suspend occurs.
Before version 8.4, the extended long busy condition is set for any PPRC primary volume that has the consistency group option enabled when it suspends. Starting with version 8.4, the extended long busy condition is set only for PPRC primary volumes that are in full duplex state when they suspend. Global Mirror or Global Copy pairs will not have a write impact that is caused by freeze events, independent of how the consistency group flag is set.
Nevertheless, different replication types should still be separated on different LSS pairs to ensure independent management of PPRC pairs that belong together in a logical session context.