Hi,
We were running Incremental forever backups and then replicating. I noticed that there were random differentials being created and asked the question of Unitrends…
"I still don’t understand why it does incrementals and differentials and then replicates them both? Why bother with the incrementals, and just set up a differential schedule? Can you please get someone to explain the theory to me, I have tried to find some docu as to why that happens but it is a bit scarce. All I see is extra space being taken up by backups (the diff’s) that I don’t schedule and have no control over.
and got the reply
Happy to answer. In the Incremental Forever strategy as you know, the backups are being scheduled on as regularly periodic incremental backups (whether it’s hourly, daily, several times daily or some other time scheme). The program decides when it’s going to ‘synthesize’ differentials and masters. Why differentials? This was originally done because when vaulting, unless it is a virtual machine, incrementals do NOT vault; only bare metals, masters and differentials. In order to get the incremental data up to the vault for that data protection, the program decides (usually within about a week to 10 days at most) to auto generate/synthesize a differential to get the data to the vault. Next, after so many days/weeks and dependent on factors (such as upgrades to the environment, how long since last master, etc.) that the system synthesizes a master and it goes to the vault.
Therefore, if the customer is feeling frustrated by seeing a bunch of non-scheduled differentials in additional to the incrementals that they are expecting, then this is how and why it happens. If they don’t want that going on, I would strongly recommend going to a full w/differentials or full w/incrementals schedule. (I personally prefer the daily differential to the incremental because once again, incrementals don’t vault and in a true DR (disaster recovery) scenario, I want my data vaulted as soon as possible. But that’s just me, “your highway mileage may vary” LOL)
So that answered that…
and then I did query that this is meant to be replication and not vaulting but the same theory applies.
So I was also a bit scared about a: having to do new masters weekly/fortnighly whatever etc and then how long it would take to replicate but got this answer
Yes the new master will replicate – but realise that it does a block level differential check against the existing master and only replicated the changed blocks (compressed and deduplicated) across the link – so across a 5Mb link you should be fine. It doesn’t send the whole image over as it doesn’t need to.
Which alleviated my fears re-markedly and after testing i found that the replication worked very fast and very well even for the masters. The only “issue” was that the masters take a little longer to process at the backup stage, but i figure you can’t have it all.
So now we run a Monthly Master that starts on a Friday night and is replicated over the weekend and differentials the rest of the time and it works just perfectly.
Hope that helps 
Cheers