Graphs. Lots and lots of graphs. If you're collecting all the relevant performance metrics for your environment, you can graph them up, look at where they're heading, and make a "squint-n-guess" estimate of where you'll hit capacity. If you want to get more scientific, you can also all sorts of fancy mathematical and statistical voodoo that will get you a result that might feel "more accurate", but given the limitations of the data and variance in the real-world operational environment, I'm not sure the extra effort is justified.
Given that you provide no indications of what environment you're in or what you've already got running, it's impossible to provide any reasonable suggestions for specific tools, but I'll just say that I've never felt a need to go any further than rrdtool. That thing is freaking magical.