Most monitoring systems are designed for exactly that - measuring the behaviour of a live system over a long period of time. There are lots of tools available for doing this - Bob's answer here gives links to some of the stuff available.
However there are a couple of problems with this.
First these are not really setup for testing purposes - you're going to have to think about how you reconcile these metrics with your testing.
Secondly, while the standard system metrics can give a good pointer to how performance is constrained - they don't actually measure transactional performance. The important thing is to deliver the data the user requests promptly - not to keep the CPU usage below some arbitrary value - so you need to think about how you measure actual end-to-end performance.
The other thing to bear in mind, is that application tuning is a very empirical art - you're going to have to colelct and visualize a lot of data - most of the monitoring tools aren't really very good for this.