During the Write phase of RecoverPoint replication, when does the write splitter send an acknowledgement back to the host that initiated the write?

The RecoverPoint write splitter is designed to be transparent to the host application and maintain local production performance. When a host initiates a write, the splitter intercepts it and performs two actions in parallel: it sends the write to the production LUN and also to the local RecoverPoint Appliance (RPA). The critical step for the host is ensuring the write is committed locally. Therefore, the splitter waits for an acknowledgement from the production storage array confirming the write has been successfully received. Only after receiving this confirmation from the production LUN does the splitter send an acknowledgement back to the host, completing the I/O operation from the host's perspective.

A. After the splitter has received an acknowledgement from the RPAs This is incorrect. Waiting for the RPA would introduce network and replication latency into the host's write path, which would severely degrade application performance.
C. As soon as the splitter receives the write This is incorrect. Acknowledging the write before it is committed to persistent storage (the production LUN) would violate write integrity and risk data loss if the storage write were to fail.
D. After the Distribution phase has completed This is incorrect. The Distribution phase is a later stage in the replication cycle managed by the RPAs. It is completely decoupled from the initial host write acknowledgement process.

1. Dell EMC RecoverPoint Concepts and Architecture Technical Notes (April 2017)

Page 13, "The write-splitting process": The document states, "The splitter sends the write to the production LUN. When the production LUN acknowledges the write, the splitter sends an acknowledgment to the host." This explicitly confirms that the host acknowledgement is dependent on the production LUN's acknowledgement.

2. Dell EMC RecoverPoint for Virtual Machines 5.2 Administrator's Guide

Page 20, "How RecoverPoint for VMs works": In the description of the I/O flow, it details that the vSphere I/O filter (the splitter component in this context) passes the write to the datastore. The guide explains, "The write is acknowledged to the guest OS on the production VM after it is written to the production datastore." This reinforces the principle across different RecoverPoint implementations.