This quote is taken from a presentation given by
Rico Mariani called
Designing .NET Class Libraries: CLR Performance Tips
People use this phrase, "Premature optimization is the root of all
evil," as an excuse not to think
about performance at all in their early planning...Say I wanted
to code a sort..so I say, "Oh, no, I'm not going to invest in my sorting
algorithm just now. That would be premature.
What I'm going to do is code it
up the easy way first and then make it fast later."... So they
write the easiest sort that they can think of: bubble sort. Great. Got
that bubble sort in there. It's nice and easy. Then later they find out
it's too slow. Big surprise there. Stunning development: BubbleSort
about the worst sort that isn't completely stupid...so they've got to
replace it. So they go pick up Quicksort. Now, let me ask you, if you
were in a position where you needed to substitute the BubbleSort
algorithm--and I'm using BubbleSort as kind of the proxy for the problem
here--If you were substituting your BubbleSort with a Quicksort, how
much learning did you get out of coding up the BubbleSort? Are you going
to just tune it up? Are you going to, "Oh, I'll just tune it up right
all day, and I'll turn it into a Quicksort"?
You're not going to tune it
up. You're going to throw it out, right? I mean there's not a
darn thing in common between those two things except for the
greater-than test. The BubbleSort is gone and pitched out the window and
you're basically writing a whole new sort from scratch, and you spent a
whole lot of time maybe, living with your BubbleSort and planning around
it or not thinking about what you need to be doing, and now at the end,
you find yourself having to discover the Quicksort...But a lot of times,
there is no off-the-shelf that you're going to plug in.
Your design is fundamentally
flawed, and you at the last minute are going to have to go back
to the drawing board and figure out how the heck you're going to do
something that's plug-compatible with the "easiest way" that is going to
fit into your overall system. You can easily find yourself doomed.
...Make sure that you have the right algorithm in place and that you
have a plan in place that lets you be substantially likely to succeed
against your performance goals. That's performance planning.
