you should use the latest stable version, that's what it's there for.
nsca pushes bulk data to a central server, nrpe goes from a central server to all the nodes. nrpe is awful, and I would personally not recommend using it. I would recommend running your checks over an existing network service, e.g. ssh which is much more secure etc. Running nrpe just opens more network holes and it generally crude and rubbish.
bandwidth is negligible. note that it obviously relates, as does server load, to how often you run these checks. there is twice as much bandwidth and cpu requirements from running a check every 5 minutes as opposed to every 10 minutes etc.
BTW, my personal opinion is that whilst it is something of a "standard" nagios is a VERY poor monitoring platform. It's just an archaic under powered web interface with a poor engine that just crudley runs external scripts on a schedule. any analysis of the monitoring data, graphing etc, requires further external packages, e.g. pnp4nagios, which integrate very very poorly into the overall environment. I would suggest using a genuine SNMP based monitoring program, e.g. opennms, zabbix, hyperic etc.
Last edited by acid_kewpie; 10-28-2011 at 04:35 AM.