Generally, we don't have to care about Puppet Master's performances when we have few nodes to manage.
Few is definitively a relative word; I would say any number lower than one hundred nodes, which varies according to various factors, such as the following:
System resources: The bare performances of the system, physical or virtual, where our Puppet Master is running are, obviously, a decisive point. Particularly needed is the CPU, which is devoured by the
puppet master
process when it compiles the catalogs for its clients and when it makes MD5 checksums of the files served via the fileserver. Memory can be a limit too while disk I/O should generally not be a bottleneck.Average number of resources for node: More resources we manage in a node, the bigger the catalog becomes, and it takes more time to compile it on Puppet Master, to deliver it via network and finally to receive and process the clients' reports.
Number of managed nodes: The more nodes we have in our infrastructure...