There are a number of ways to do this -

5+ years ago Mavenspire open sourced a script that basically implemented the following workflow (in the powershell and compcu java command line)

1) Take a replay of the destination replications in DR (so that we can read/write to the data without endangering the active replication)

2) Mount each volume to the backup server. Ensure that the backup server had the software modules to mount VMFS, EXT, and whatever else needed so that the volumes could mount.

3) Run a backup job on the volume mounted (differential) as push to tape.

Pros:

1) Fast backup since the LUNS are local to the backup server

2) Worked for basically everything except open file applications - but worked fine for their backup dump files

3) Could be done during the day vs overnight since its based on snapshots

Cons:

1) The scripting / scheduling was a little complicated with 100TB of data and slow tapes

2)  Scripts had to be updated as new things were added to the environment

3) Didnt work well for OS/bare metal backups - that still needs to be done with agents