Benchmarks

This chapter gives some benchmark results that compare the performance of the Str class, char* strings and the standard C++ string class in a variety of common string operations. Remember that a benchmark of this type represents a "snapshot" view of a complex system that contains many variables. Variables that can influence the result include:

• Processor, Cache Sizes, Memory Speeds, etc.: These benchmarks were run using an AMD64 3200+ with 64k of L1 cache, 512K L2 cache, and 1 Gb of DDR400 DRAM.
• Compiler Version and Settings: The compiler was gcc 3.3.5. The only compiler flag used was -O2
• Optimization Variability: The gcc optimizer does an excellent job of converting short char* functions into optimized code, how well this optimization scales to more complicated "real life" situations is less clear.
• C Library Version: The version used was glibc-2.3.4
• Types of tests: Different test types will bring out different strengths / weaknesses of each approach. For example, it is possible to make char* appear to have poor performance by over-emphasizing operations that are $O(n)$ when using char* but $O(1)$ under Str. The approach here was to choose simple, commonly used operations for the benchmarks.
• Coding Style: Often more specialized approaches to a given problem will yield better results but these same approaches can not be generalized to handle a wider range of problems. For example, when appending substrings together, performance can be measurably improved if the maximum length of the final string is known ahead of time. These tests try to keep to general approaches.

The benchmarking code used is included with Str under the benchmark directory. On to the benchmarks...