If you're considering data deduplication/disk-based backup, the fundamental starting point should be RPO (recovery point objective) and RTO (recovery time objective). If RPO's are very tight and you want to send dedupe data over the network - then you'll want to consider server-based solutions like Avamar, which are more expensive.

If on the other hand your applications have more relaxed RPO requirements then you'll have more time to de-dupe and send de-dupe data over the net. This approach will save a bundle.

Another consideration is size and scale. If you're small (less than 10 tb) then there are a variety of vendor/product choices. If greater than 10TB, the choices narrow significantly.

The other consideration is doing in-line or post process dedupe. Data domain and Diligent, etc do inline, Falconstor and others do post process.

From a business perspective, post process allows you to do the backup faster, then you perform the dedupe. But you've added another process in the stream meaning the whole procedure takes longer and is more complex. Meaning RTO will be extended.

In line slows down the backup process but overall it's faster and easier to manage. So your RPO goes up but your RTO goes down.

You might also want to consider outsourcing disk-based backup as a remote service (e.g. IBM and others) which brings other considerations.