Chapter 10, Local Buffer Optimization

1. Micro-benchmarks can measure the performance of small fragments of code in isolation. To measure the performance of the same fragment in the context of a program, we have to use a profiler.
2. Processing small amounts of data usually involve a correspondingly small amount of computing and are therefore very fast. Memory allocation adds a constant overhead, not proportional to the data size. The relative impact is larger when the processing time is short. In addition, memory allocation may use a global lock or otherwise serialize multiple threads.
3. Local buffer optimization replaces external memory allocation with a buffer that is a part of the object itself. This avoids the cost, and the overhead, of an additional memory allocation.
4. The object has to be constructed and the memory for it must be allocated, regardless of whether any secondary allocations happen. This allocation has some cost – more if the object is allocated on...