How to Make Things Faster

Chapter 35. The Bottleneck

Maybe you’ve heard that performance is all about eliminating bottlenecks. The word itself is a metaphor: it doesn’t matter how big the bottle’s body is, the amount you can get out of it in a given time interval is constricted by the size of the little opening there at the end.

Stories about bottles and water pipes and such are fine for learning principles, but they don’t tell you what to type. And it so happens that you have to be careful about how you define bottleneck for the principle to be helpful. Here’s the definition you need: the resource that dominates the response time for a task execution is that execution’s bottleneck. If you have a profile for the execution you’re observing, then the definition is even simpler:

The bottleneck for a program execution is that execution’s largest response time contributor.

Let’s try it out. In the Payroll story, what’s the bottleneck of the PYUGEN execution? Here’s the profile again, to jog your memory:

	Duration
Subroutine	seconds	%	Count	Mean
SQL*Net message from client	984.010	50.3%	95,161	0.010 340
SQL*Net more data from client	418.820	21.4%	3,345	0.125 208
db file sequential read	279.340	14.3%	45,084	0.006 196
CPU service, EXEC calls	136.880	7.0%	67,888	0.002 016
CPU service, PARSE calls	74.490	3.8%	10,098	0.007 377
CPU service, FETCH calls	37.320	1.9%	57,217	0.000 652
latch free	23.690	1.2%	34,695	0.000 683
8 other subroutines	2.920	0.1%	111,829	0.000 026
Total (15)	1,957.470 ...