Hi Bruno,

The general rule of thumb is to create all your Cassandra nodes in one datacenter and point all your DataMiner agents to that datacenter. This mainly to keep the latency between the Cassandra nodes as low as possible. If you are seeing that some DataMiner nodes are too far away (typically when the latency is >50ms) from that datacenter, then you can spin up a new datacenter within Cassandra that is closer to those DataMiner agents that are far away. Keep in mind that all data will be available in every datacenter of Cassandra with its own replication factor configured for that datacenter, so the cost of spinning up a new datacenter can be high if you have a lot of data. You can play around with the replication factor on the different locations (how many copies of the data are kept in the datacenters) to keep the cost under control.

If you have a lot of DataMiner agents spread around the world (high latency), then it can be very costly to spin up all those datacenters (containing all data) over the world. Then it might be worth considering having separate DataMiner systems for every region and letting them push information to a central DMS where your NOC will connect. The disadvantage here is that you will not be able to see the detailed/all information of every element (device) in the central DMS, but if those regional DMS's are reachable for the NOC you could work with links to connect to the other DMS's when needed. In such a setup it might also be worthwhile to look into third-party authentication system (e.g. through SAML) for single sign-on.