IMHO test and keep track of performance
start off with two servers, SQL on one, NPMNTA on the other. see how that goes to see if you need to deploy 50% more HW
my environment is 100% virtualized. that allows me to place each in server where i want it
our servers have no HD. all storage comes off a 4 tier SAN. if you go that way , ensure that your SAN cam meet both your peak and steady state IPOS
The NTA 4 FAQ discusses some recommendations - looks like you've read this?
The optimal setup would be 3 separate servers as per the reqs outlined above. Based on the info you've given, it looks like you are going to be using netflow extensively and receiving an average of 40k flows per sec. You do not mention how many nodes/interfaces you are polling with NPM, or your license sizes?
If you want to reduce the number of servers and you are not polling a lot of elements with NPM, then you could combine your Poller and SQL DB on the same server. This would not have been recommended prior to NTA v4, however now that NTA 4 uses its own flow storage database, it's important that it has a sever of its own particularly in your environment given the flows per sec you quoted.
thank you for the answer. Now it becomes clearly. So if I have a lot of Interfaces to poll with NPM, I need a performant SQL Server. Also if I use the new FSDB on an additional server. The third server I need then for the NPM/NTA Poller.
We have 34 Nodes with about 1000 Interfaces.
Just to add on what Ciaran stated.
The key take away is to have your FSDB and SQL server on separate servers. If you have a small deployment, which is sounds like you don't, you could get away with a single box solution. NTA 4.X smallest recommended foot print is 2 servers. One hosting your NPM/NTA/Orion SQL DB and one server to host your FSDB.