I have a munin-master running on an Atom C2338 (very low-end), which monitors about 20 hosts. The whole system uses about 300MB RAM and 40% of a CPU core. It's

I have a munin-master running on an Atom C2338 (very low-end), which monitors about 20 hosts. The whole system uses about 300MB RAM and 40% of a CPU core. It's a little intensive because it recomputes half the images every 5 min, when the cron runs (that's configurable). In terms of disk, munin uses 750MB, and it's not growing thanks to rrdtools; and it writes 1MB/s on average. Bandwidth usage is 5Mb/s on average (to get data from all munin nodes, also every 5 min in my config) Munin nodes themselves are extremely lightweight. Essentially, the master calls the munin-node daemon via a TCP socket, which starts a plugin process and pipes its stdout to the socket. Plugins usually simply call a command like "df" or "netstat", parse its output, and write it in munin's format.