How to Make Things Faster

Chapter 88. When Ratios Are Reliable

Ratios become reliable metrics for comparing two systems (or two system states) if you first eliminate all the waste from the ratios you are comparing. In the Oracle story, if you focus first on eliminating wasteful buffer cache accesses, then you’ll never get tricked by the evil genie who padded your numerator and denominator by almost a million apiece to give you a 0.999 hit rate.¹ In the mpg story, if you first eliminate the wasteful distance, then you’ll never get tricked into thinking that the better mpg route is the better overall route.

But identifying waste isn’t always as simple as it sounds. Let’s examine that for a minute, going back to the mpg story. Imagine there’s a third route called C that is six miles long, on which you can get 30 mpg. You’ve already seen that you can’t just say route C is superior because of its better mpg score. But is it possible that this new longer route C could be better than the shorter route A? Let’s see:

Measure	Route A	Route C	Growth factor
Distance (miles)	5	6	1.2
Fuel burned (gallons)	0.25	0.20	0.8
Duration (minutes)	10	8	0.8
Fuel mileage (mpg)	20	30	1.5

This route C is indeed an improvement over route A in two important regards: even though the route is longer, it gets you to work 2 minutes sooner, and it burns 0.05 gallons less fuel. Is it tire and engine wear that you’re concerned about? Route C is slightly longer than route A, but the highway driving may actually be a little ...