First you have to figure out whether you want to use the cloud or not. It's not mandatory. You can just have the UMH running on a VM within your plant network. If you still want data in the cloud, you can ingest data from the local TimescaleDB instance to a (PostgreSQL) database in the cloud. Going fully for the cloud is more tricky, unless your data already lives there. In that case you would indeed deploy the UMH on an Azure VM. But then how would you connect all your on-premises systems with that cloud VM? You could expose your Grafana dashboards to the internet but you have to be very careful how you configure your identity and access management (you don't want to share your insights with the whole world 😉 ). If I would be architecting a UNS like the UMH I would personally opt for having the UNS run on-premises, and just back up the historian's data to a cloud database. This way you have fast data on-premises and all of the historical data in the cloud for advanced analytics. I think @Jermuk can share his experience on this matter as well

I would connect all my on-prem machines via a HiveMQ broker and then ingest the data from the hiveMQ broker with node-red in UMH. (move data from external hivemq broker to the UMH internal one to be further processed)

Yes, for multi site one UMH installation per site would be the way to go. Then one "master" UMH would unite the installations on each site

When you do this would the data structure be replicated everywhere or just what you need at each site then everything comes together in the master?

Every site has then only the tags and the structure that is relevant for the site, e.g. umh.v1.enterpriseName.siteName...... if there is a site called "ABC" but with tags in the topic "umh.v1.enterpriseName.AAA", there will be in the future also a warning about this. Then the master node, which is probably marked as "enterpriseName", receives data from all sites and has access to everything

This makes sense! I was wondering about SAP data which I assume we would connect to the master node, and would send from the master to the site node but I guess even in that case it will only be that site’s ERP data

you could also do it like this, if the SAP system is in the cloud