Each cluster can be both a replication source and a target, allowing customers to split applications between regionally separate clusters while protecting application availability at each location.
The volume mapping on the source SDC sends writes for replicated data to the SDR, which duplicates the write and forwards it. The local SDSs process writes normally, while the SDR assembles the journal files that contain checkpoints to preserve write order.
Journals are batched in the journal buffer on the source system. As they near the head of the queue, they are scanned, and duplicate block writes are consolidated (folded) to minimize the volume of data being sent over the wire.
The journal intervals are sent to the remote target journal buffer by the SDR over dedicated subnets on local networks or external WAN networks assigned to replication. Once the intervals are acknowledged at the target journal, the SDR removes the intervals from the source.
On the target system, the SDR processes the journals and passes the writes on to the relevant SDSs. The SDSs manage the primary and secondary copies as usual. You may be wondering: How is compression affected in replicated volumes? The short answer is that compressed data is not sent over the WAN. The SDR is a mediator between the SDC and the SDS, and it plays no role in compression. The SDS is responsible for compressing writes and storing them to disks local to the host on which it runs.
Once the target-side SDR receives acknowledgment from the target-side SDS, it goes to the next write contained in the journal interval being processed. When the last write in a journal interval is processed and acknowledged, the interval is deleted, and the journal capacity is made available for reuse.
Several other SDR subprocesses work together to protect the integrity of your data, but this description addresses all the fundamentals.
One limitation worthy of mention relates to volume migration. Migrating replicated volumes from one protection domain to another is not possible. This limitation is because the replication journals do not span protection domains.