Avoid the Top 2 Performance Pitfalls: Logging and API Design

At the 2016 QCon conference, Martin Thompson, founder and CTO of LMAX, presented "Top 10 Performance Mistakes" and this summary focuses on the first two items.

1. Logging

Thompson argues that logging is the most common performance bottleneck. He shows a chart (see image) where the number of logging threads grows linearly with total execution time.

Testing across many logging frameworks produced the same pattern, indicating that loggers are a major performance bottleneck. Thompson recommends using asynchronous loggers and deduplicating log messages: record a message only the first time it occurs, then use a timer for subsequent repetitions.

2. API Design

Many API signatures are poorly designed from a performance perspective. Consider the following method: public String[] split(String regex) The fixed‑size array return forces the implementation to allocate a temporary structure and copy data, and callers must copy the array again if they need to sort or further process the results.

A better design is to return an Iterable instead of a fixed array: public Iterable<String> split(String regex) For even higher performance, the method can be changed to accept a destination collection supplied by the caller, eliminating the need for intermediate copies:

public void split(String regex, Collection<String> dst)

This allows callers to choose the most appropriate collection type—e.g., a Set to remove duplicates or a TreeMap for ordered results—thereby improving both memory usage and execution speed.