Profiling

If benchmarking is discovering how long a piece of code takes to run as a whole, then profiling means figuring out how much time each function, method, or instruction spends. The primary purpose of using a profiler is to optimize one's code. You don't want to go tweaking and rewriting your code blindly, spending time on optimizations that might not even matter. Profiling gives you precisely that. It identifies who are the primary contributors to your program's execution time. Once you know them, you can focus on revising them first, as it will have the biggest effect.

Tracing and sampling profilers

A sampling profiler (also known as a statistical profiler) works by regularly taking a stack snapshot for each live thread and recording which methods were on the stack at that moment. Then, it uses this information to infer approximately how long each method has been running. The reasoning is the following: if 9 of 10 stack samples contain method foo(), and only 1 sample contains bar(), then foo must run 9 times longer than bar.

Because it is based on statistics, a sampling profiler doesn't really know how many invocations of each method happened or how much time each method spent exactly. But, surprisingly, statistical profilers are quite accurate if you give them a reasonable sampling rate (around 100-10000 Hz) and some time to run.

A tracing profiler is another name for an instrumenting profiler. Instrumentation means that the profiler is going to inject its profiling code into every method of the program. The profiling code will track every entrance and exit in/out of the method and the time spent inside. This type of profiling makes no guesses. It knows for sure what methods were the slowest and invoked the most number of times.

In the case of Clojure, the instrumentation of every method poses a huge problem. Because of the way the Clojure compiler is implemented, a bare Clojure application contains thousands of classes, let alone methods. Setting an instrumenting profiler on a Clojure program will often grind your program to a halt. It will take 5-10 minutes for the profiler to finish instrumenting, and good luck using your program after that. However, to be fair, some JVM profilers have quite an efficient instrumentation mechanism.

But even if an instrumenting profiler doesn't render your program completely unusable, there is another problem with these kinds of profilers. Because instrumentation is, by definition, invasive (methods are recompiled to contain code that you didn't write), the veracity of its results dramatically diminishes. Furthermore, depending on the implementation, an instrumenting profiler can disable JIT and inlining of your code for better introspection. Suddenly, the instrumented code that you profile stops resembling the original code entirely. In such a scenario, an instrumenting profiler becomes much less precise than a sampling one.

Therefore, it is advised to use a sampling profiler to profile Clojure programs or pick the sampling profiling option in profilers that support both mechanisms.