CDP

We’ve implemented it at a customer’s site when very low RTO / RPO requirements for their ERP database were needed. Local setup with 10Gbps connection between hosts with direct attached storage and a decent amount of memory for the proxies does the job.

I don’t use it in my environment personally. Very niche feature IMO. The only caveats to its implementation I hear rumblings on is installing the I/O driver on the Hosts and getting it to work well. But once implemented, I don’t hear much negativity.

I’m using it on a few test VM’s right now and it’s working pretty well. I may implement it larger scale shortly.

I’m a big SRM shop here so it would be nice to save a few bucks, but SRM with SAN replication works REALLY REALLY well and has been solid for years so I want to be sure before I make the jump.  I”m testing a few replicas as well for things that don’t need such a tight RPO/RTO.

What's the most VM’s any of you guys have run with CDP? What NW bandwidth do you have as well? 

I have it running in production replicating 3 VM’s.  There are only 3 caveats I’ve noticed. 

1. The IO Filter is one which has been mentioned.  Installing hasn’t been an issue, but uninstalling or updating has.  Often it wants to put my hosts into maintenance to do this work, but because I have a 2-host cluster, its not always easily done.  That said, you can manually uninstall and Veeam will catch up with it.  I haven’t updated my filter drivers.  I also posted an entry a while back where I was having some strange errors and it turned out that I didn’t have the same version of IO Filter on each end.
2. Prepare for more space to be utilized on the target storage array.  My current storage array, to be replaced soon has run out of space a couple of times because CDP with the occasional checkpoint does consume more space than snapshot-based replication.  With snapshot replica’s, if you’re getting low, it will just not run and throw an error.  With CDP, it will take you down to nearly zero (into the MB) and run the datastore completely out of space...which can be troublesome to resolve, especially if you have other VM’s on the same datastore, especially if those VM’s are live (mine weren’t live buy I can see the potential for issues here).
3. Make sure you have your policy configured to be compatible with your IO changes on the VM’s and what your WAN link can handle.  It does require some tuning to make sure things keep up, as well as some alerting tweaks so that you’re not getting blasted with notifications if you fall too far outside of your RPO.  I could have been more conservative, but I wanted to try and take it to the edge of what we could do, and I got slammed with emails on the regular.  I did tweak how far outside of my RPO I could get before getting alerted and that helped, but I think I eventually did scale back the RPO from second to something like 15 minutes or so.