You need more information than just “5000 devices”.
For example, are these “devices” servers which can run the Icinga Agent
locally, or are they routers, switches and similar things which need another
machine to get data from them by, for example, SNMP?
Also, how many service checks do you need to run on each device? I mean
things like disk space, process numbers, load average, network traffic…
Finally, how frequently do you need updates of the data (how long can you
accept between service checks, when you won’t know there’s a problem if one is
Oh, and where are these “devices”? All on a local high-speed network, or
widely distributed across the world on some fast, some slow, network links, or
in a few good-bandwidth data centres?
All of those things will play a role in determining:
a) what architecture you need to set up (one or two masters, zero or more
b) what sort of spec you need for the machines involved.