Hey Toby.  Congrats on finally expanding!  We have a separate node for running background refreshes and while it seems to be CPU heavy and disk heavy, it doesn't seem to take up much memory at all, in fact over the past week, RAM max peaked at 17 gig.  I think the vizql is the one that takes up the most RAM.  We have 4 backgrounders actively running, but intend to increase it over time...

Are you setting up for high availability, Toby? If so, you'll need a third "observer" host added to the cluster. It doesn't need to do anything, just run the bare minimum of processes (won't consume core licenses, this way). All it would be there for is breaking a tie if both nodes report that the other is down.

I might start out simply by mirroring what you had on the single instance, assuming that will get you to the number of backgrounders you need. Combined with the "dummy" observer host, you'd have full HA that way, along with the increased capacity.

If you are constantly running lots of extracts, consider dedicating a full host to that work, as it can interfere with the user experience if you refresh too many extracts at the same time on a host responsible for running the application server, vizql, etc.

What is the up-ratio for the coordination service before it reports that the cluster is down?

sorry about that.  With a high availability cluster, there is some notion that if a certain % of nodes are down, then Tableau Server can continue to function perfectly fine.  But if it falls below the threshold, then Tableau Server as a whole goes down.  I just can't recall what this ratio is, but it's a cert question. and I couldn't find the answer in the admin guide

Install drivers, make sure host OS is set up the same as your Primary. If you want an extra assurance of working-ness (that's a technical term), you could install Tableau Server as a standalone on your new host, test it out and make sure it works properly, then do a clean uninstall on the new host. Install the Worker software, then stop Server on the primary, add the new worker to the cluster, and start again. That's it in a nutshell.

Yes, that's fine. Just make sure you install the drivers beforehand. If you're doing a dedicated Backgrounder host, then that's pretty simple, right? 8 is a good number to start. Just throw 8 of them on there and remove whatever backgrounder processes you've allocated to your Primary. You can tweak things if necessary later. If you have per-process CPU / memory data for your existing Backgrounders, you can guesstimate what kind of resources they'll consume. For us, RAM has only been a concern during really high-volume Subscription processing times. We were running about 12 backgrounders on a 16-core box, and we'd hit our 64 GB limit during the 6am hour sometimes since that's when everyone wanted their Subscriptions sent.