How to Make Things Faster

Chapter 58. Reaching the End of Your Data

The how-long–why loop is a tidy way to think about performance. Every response time problem, at its highest level of analysis, is a how-long problem. How long did PYUGEN take in the Payroll story? 1,957 seconds. Likewise, every price problem, at its highest level of analysis, is a how-much problem. How much did your bag of groceries cost? $55.48. In both worlds, the next question you’ll ask is why.

Answering the why question requires an analytical plunge, a drill-down that you can accomplish only if you have more detailed data to drill into. If you know that PYUGEN took 1,957 seconds only because you timed it with a stopwatch, then you’re not going to be able to answer the why question until you trace. Likewise, if all you see is the $55.48 line item on your credit card statement, you won’t be able to answer why until you find your receipt.

Every how-long question stimulates its own why. The process is recursive, because each new why inspires a new, more detailed how-long (or how-much) question that can inspire a new why question, like this:

	The users are upset.	Mom’s upset.
Why?	Because Payroll is taking too long	Because the tomatoes cost too much
How long/much?	1,957 seconds	$15.48
Why?	Because 984 sec (50%) is network I/O calls	Because $15.48 (100%) is tomatoes (after returning the chocolate bars)
How long/much?	95,161 calls costing 10 msec per call	1.2 lb of tomatoes costing $12.90/lb
Why?	That detail is not available ...