Point taken on the configurability. I think a strong argument can be made for the number of cores as being the default for cpu_num. However, it's also quite sensible to have the other two measures: number of physical CPUs, and number of virtual CPUs (what gmond currently reports).
How about, instead of cpu_num, reporting: cpu_count_cores cpu_count_physical cpu_count_virtual They are all very easily calculated based on the contents of /proc/cpuinfo under Linux. I am not as certain how one would go about calculating all three in other environments... Given that one had all 3, one could make the load calculation configurable thus: metric { name = "load_one" value_threshold = "1.0" basis = "cpu_count_cores" } Comments?