Thin to Thin and Any to Any
May 19th, 2008
Hitachi Dynamic Provisioning, HDP, has been enhanced today to support Synchronous (TrueCopy) and Asynchronous (HUR) remote replication. Now we can replicate thin provisioned volumes to thin provisioned volumes, on any virtualized storage device to any virtualized storage device. What does this all mean?
Open systems users must reserve capacity for a volume before they can start writing to the volume. They don’t know how much they need before they start writing, but they do know that they don’t want to run out of space, since an out of space condition would be disruptive to the application. Therefore a standard practice has been to reserve, or allocate, more capacity then would ever be required, some times by a factor of two or more. This is justified because the cost of storage capacity is thought to be cheaper than the operational cost of recovering and fixing an out of space condition.
Unfortunately this allocated, unused capacity is multiplied many times over when the primary production volume is copied or replicated for backup cycles, development test, extract/translate/load, data mining, and distance replication. Since a control unit makes copies by copying blocks, it does not know how much of the volume actually contains valid data, so it copies all the blocks that were allocated. All that allocated unused space that is copied many times over not only wastes storage capacity along with power, cooling, and space; it also wastes time, processing cycles, and other storage resources for data that is not there. The resources that are expended for remote replication is especially expensive, since it requires cache resources, journal disks, and expensive remote links, as well as a control unit on each end.
Hitachi Dynamic Provisioning still allows a user to request an allocation for more capacity than he needs, but the capacity that is actually allocated to the volume is done in 42 MB pages as the user begins to write to his volume. In this way we ‘thin” provision the volume and eliminate the waste of allocated unused space. Since the control unit is allocating these pages, when it is asked to make a copy of this volume, it only copies the pages that were allocated and not the whole volume. This not only saves the waste of allocated unused capacity, it also saves the other resources that would normally be consumed to process that additional capacity.
In the case of remote replication, if we only process the allocated pages, we may cut the replication time down by 50% or more! We can also replicate the thin provisioned volume to another thin provisioned volume at the remote site. Then if we have to use the remote replica for disaster recovery, the application can be restarted at the remote site, and continue to grow the volume up to its original allocation request.
Since the USP V/VM control unit that supports HDP can also support virtualization of heterogeneous storage systems, the HDP pool of thin provisioned volumes can reside on any virtualized storage system and be replicated to any other storage system that is virtualized behind a remote USP V/VM with HDP.
In this way HDS provides thin to thin and any to any replication, saving capacity, power, cooling, space, time, and other operational resources.
