MRTG will do much more than graph traffic:
information can be collected either via SNMP or from scripts, and is stored in such a way that the latest information is very detailed, and the data is averaged more and more to allow for graphing of data over longer periods of time without using huge amounts of disk space to store the raw data. In addition to the graphging function, it has the ability to run scripts on triggering thresholds - allowing for everything from emails/sms to be sent right through to automated shutdowns.
The only limitation of MTRG is that it it will only display 2 sets of data on any graph. Because of this, Tobias then created the rrdtool - http://oss.oetiker.ch/rrdtool/ which didn't have this limitation. Many packages - for example http://munin.projects.linpro.no, Cacti, Cricket are build on top of the original rrdtool engine.
This is one of the most useful tools that any systems administrator can implement, as it will eventually be telling you when to look at something before it goes wrong, allowing for a proactive approach to administreing your server(s).