Besides ruby-prof and and the core Benchmark class, what do you use to profile your Ruby code? In particular, how do you find the bottlenecks in your code? It almost feels like I need to work on my own little tool just to figure out where all the time is being spent in my code.
I realize ruby-prof provides this, but the output is frankly very confusing and doesn’t make it easy to find out which actual blocks of your own code are the source of the issue (it tells you about which method calls took the most time though). So I’m not really getting as much out of it as I’d like, and haven’t really been able to make use of it.
Perhaps I’m doing it wrong? Are there alternatives? Google searches don’t bring up anything for me.
Lots of profilers are like that.
What you need to know is not where the program spends its time, but why. Any references on Dynamic Code Analysis?
ADDED: Here’s how I find “bottlenecks” in my code. (I hate that word.)
Here’s a list of reasons why.
It is perfectly natural to assume that to find “bottlenecks” you have to somehow do a lot of measuring.
It is so natural that nearly all profilers are based on it.
Actually, finding and measuring are not the same problem. Measuring is needed to see if what you found (and fixed) made a difference. Finding what to fix, to me, is more like debugging than measuring.
The simplest way to explain it is to start from an infinite, or nearly infinite loop. How do you find it? You pause it and look at the stack, right? because you know the problem is somewhere on the stack. You only need to pause it once, and then you need to study the code on the stack. Pause it a few times if you want to be sure you’ve found it.
Suppose the code only takes twice as long as necessary. That means when you pause it, there is a 50% chance you will see it doing the unnecessary thing. If you pause it and look at it 10 times, you will catch it in the act roughly 5 times. In fact, as soon as you see it doing something you can optimize on as few as 2 samples, you’ve found a “bottleneck”. Fix it, measure the speedup, show it off, and repeat.
Even if your biggest problem is not very big, this method will eventually find it.
Also, there’s a magnification phenomenon, where small problems become easier to find after you’ve removed larger ones. That allows you to keep going until the code is nearly optimal.
P.S. After you’ve done this, there may still be opportunities for speedup. For example, optimization algorithms can depend on numerical stability. Message-driven architectures can make it harder to trace why code is being executed. In real-time software, a performance problem may only happen occasionally, and be less easy to sample. This requires more cleverness. Falling back on just measuring doesn’t do it.