HP StorageWorks SAN Virtualization Services Platform Best Practices Guide (5697-0935, May 2011)

10 Synchronous mirroring
Synchronous mirroring is advantageous because the distance between sites can reach 100 KM
(in a stretched domain), failover is transparent for most events, and the RPO and RTO are
predictable.
Overview
A synchronous mirror is started by creating a synchronous mirror group for the virtual disk to be
mirrored. This group replaces the virtual disk and creates an initial task representing the operations
to the source virtual disk. The hosts that have access to the primary virtual disk performs all I/O
on the synchronous mirror group instead of the virtual disk. This change is transparent to the host.
A group has one task when it is created. Tasks may be added to create additional copies, up to
the number currently supported. Each write is performed on both the primary virtual disk and
secondary virtual disk tasks before status is returned to the host. Each read is executed to the current
primary task.
When an additional synchronous mirror task is created, a full copy of the primary virtual disk is
made to a secondary virtual disk (referred to as being in a Snap-Sync state). A synchronous mirror
copy is an ongoing process that directs host I/O to one virtual disk. After the task and the first
copy are created, additional copy tasks can be added. The additional copy task allocates a new
(secondary) virtual disk from its specified storage pool and makes it a mirror of the primary virtual
disk. The capacity of the new (secondary) virtual disk is the same as the capacity of the original
primary virtual disk.
NOTE:
• Failures may cause the "primary" task to move from one task to the other. It is not always
fixed to the original source.
• Only eight synchronous mirror tasks can make progress in a Snap-Sync state at any time. This
is the throttling mechanism used to prevent the resynchronization operation from consuming
all resources and impacting host I/O within SVSP.
Synchronous mirror metadata
In order to manage synchronous mirrors, SVSP needs a place to store state information about each
group and its tasks. This information is called metadata, and when a group is created, SVSP creates
two auxiliary internal virtual disks for each task in the group:
• State (ST) virtual disk—Contains basic information about the synchronous mirror group and
its tasks, such as the name of the group and the number of tasks in the group. The state virtual
disk also holds the dirty region map. One megabyte is allocated for the state virtual disk.
• Reallocation table (RT) virtual disk—Contains information that enables another mirror virtual
disk to be restored quickly in case of a task error. The journaling table is kept in this auxiliary
virtual disk. One megabyte is allotted for the reallocation table virtual disk size.
The capacity for this synchronous mirror metadata must come from the same storage pools as the
task's virtual disk. This has the advantage that failure of the pool affects the synchronous mirror
and its metadata. Once the pool is operational the mirror resumes. It is possible for individual
drive failures within the pool to affect some but not all of the virtual disks associated with the group,
but a failure to any one can impact the mirror.
40 Synchronous mirroring