How to Make Things Faster

Chapter 66. Predicting with Profiles

Every what-if question is an appeal for a prediction. In Richard’s story, what if you eliminate the statement that consumes 97 seconds? You should see 97 seconds less response time. What if you eliminate 800 parse calls? You’ll spend almost 8 seconds less time parsing. Without a profile—armed, say, only with information that has been aggregated across lots of different programs—trying to answer simple questions like these can be maddening. But with a profile, answering such questions is not that difficult, even for people without a lot of experience.

The fastest way to become a better predictor is to use feedback:

1. Predict something. Try your best.

2. Execute and measure the phenomenon you tried to predict, and compare the measurements to your prediction. Learn which aspects of your prediction were correct and which were incorrect, so you can incorporate what you learn into your next prediction. Go to step 1.

The key is to predict results you can actually measure, so that you can compare your predictions head-to-head against what really happened. With enough practice, your predictions will become more and more reliable.

The fastest way to become a better predictor is to practice, with feedback.