Making informed design decisions

It is not only when making decisions about trade-offs that we have to stand of the firm foundation of good performance data. After all, how can we make decisions about designing data structures for efficient memory access if we do not know how much it costs to access data in a cache-optimal order as opposed to some random order? This comes back to the first rule of performance, which you should have memorized by now: never guess about performance. This is easier said than done if our program exists as a scattering of design diagrams on a whiteboard.

You can't run a design, so how do you get measurements to guide and back up your design decisions? Some of the knowledge comes with experience. By this, I don't mean the kind of experience that says, "we have always done it this way." But you may have designed and implemented similar components and other parts of the new system. If they are reusable, they come with reliable performance...