Make sure that there is an up-to-date backup of the servers and database schemas that will be modified.

Make sure that the Content Management Server is not already in multi-master mode by verifying that the property cap.server.multipleLiveServers is set to false.

Using the information gained during the test phase, install new instances of all clients that need updating. Do not yet start the new clients, but make sure that they are correctly prepared for start-up and connection to the main servers.

Stop all editorial work.

Stop all writing clients, including, but not limited to, importers and scripts. (Make sure to reschedule all periodic tasks that would happen in the next hours.)

Shut down the Workflow Server.

Shut down the preview CAE and all other reading clients that access the Content Management Server.

Make sure that all Replication Live Servers have caught up with the Master Live Server. You may use the tool cm events on the Replication Live Servers for this purpose.

Stop all clients that are attached to the Live Servers, including, but not limited to, CAEs. At this point of time, your live site is down.

Update the folder structure of the Content Management Server. Publish the changes. (This is typically done using the same script that was used to set up the test environment.)

Make sure that all Replication Live Servers have caught up with the Master Live Server, replicating the change of the repository structure.

Start the new CAE instances that were previously installed in step 2. At this point of time, your live site is up.

Because the CAEs are starting up with empty caches, expect high load on the servers at this point of time. Wait until the load has returned to normal levels.

Start new instances of other reading clients on the Replication Live Servers.

Shutdown the Content Management Server.

Run cm multisiteconverter in the directory where the Content Management Server Tools are installed and on the host where the Content Management Server is supposed to be running. The runtime may vary depending on the actual size of the repository and the performance of the database, but 500 resources per second were achieved in a large example migration.

Change the property cap.server.multipleLiveServers to true.

Start the Content Management Server.

Update the publisher.target to include the new top-level folder as its single publication target. To this end, you will have to comment out existing configuration parameters and add new configuration parameters.

Start the workflow server.

Start the new instances of clients at the Content Management Server side. Make sure to keep track of the correct start-up order, if that is required.

Enable periodic tasks, possibly rescheduling next runs as needed based on the actual downtime.

Resume editorial work.