“Your app is slow!”
What are you going to do about it? Is it that new code the intern landed? Is it the operating system’s fault? Misconfigured hardware? Or even, god forbid, a bad algorithm?
Your thoughts run to that bloated subsystem. You know the one. It’s complicated. You’ve had your eye on it for a while. It would be so sweet to refactor it. Take some of those algorithms down to O(log n) and add some caching. Get it running really sweet. All that code. It’s got to be it.
So you spend all day and a long night crunching on it. Getting it all working right. You spend some time tracking down bugs and adding a few previously unneeded features. You get it all working, check it in, and sleep till noon. You head back into the office to be greeted by… silence.
“Isn’t it faster now?” “Nope – still slow!”
What went wrong here?
One key got skipped: measuring.
Your very first instinct whenever you see a performance issue should be to quantify it. The most powerful lines of code you can write are these:
var startTime = Platform.getTime();
// ... the code which is the problem area...
trace("Elapsed time: " + (Platform.getTime() - startTime));
I’ll often start at the root of my application and manually add blocks like these until I can identify the specific section which is problematic.
There are lots of fun variations on this idea – for instance, having a time threshold under which no output is printed to cut down on spam, or keeping track of averages/min/max, or even automatically profiling every function in your codebase (my game engine, Loom, does this – try profilerEnable and profilerDump in the Loom console!).
If you can’t nail the problem down to the point where you can measure it, it means you don’t understand the problem and you’re doomed to thrashing around with random chances until you fix it… or you just convinced yourself it’s better without changing anything. (You did all that work, didn’t you?)
The Psychological Barrier
Humans are actually VERY perceptive about fast changes in their environments. VR helmets induce motion sickness if they are more than a dozen milliseconds behind the head’s actual motion. Hearing is dependent on detecting changes in movement that are far less than a millisecond in duration.
So when you’re working on performance, eyeballing it should be plenty good, right?
Not true! People aren’t very good at remembering a specific arbitrary interval (say – 150ms vs. 100ms) and recalling it later. So when you’re optimizing load time and you shave 10% off of your 3 second load, you might not even notice that it’s faster. If you’re tired and grumpy, or just distracted, you might even think it’s slower!
It’s much easier and more reliable to just time it and keep some simple notes on your changes and what they did to your metric.
The first thing to do when you hit a performance bump is to identify and measure it… then take a deep breath and think about what might be causing it. If you can narrow the problem down to a specific hot spot, you’re golden. (Assuming you can speed that part up – but that’s why we have fancy CS degrees, right? ;))
Now all you have to do is iterate. Try something. Try ten things. Measure after each one. Make hypotheses about what might be slow and try to remove it from the equation to see if it really IS your bottleneck. There’s a whole science here – check out my book on video game optimization for a full discussion – but once you are able to measure progress you will be able to move forward.
This is an area where I often see haste lead developers into wasted hours or days or increased technical debt – when a little care and patience would crack the problem right away! So remember to take a deep breath and measure BEFORE you code. 🙂