Performance measurements for OpenMP constructs

[MarcusKlik]

Timing measurements with microsecond accuracy are needed to analyse the performance of OpenMP parallel constructs in the core code of fst. We need to determine the speedup due to parallel processing, the effect of locks in the code (e.g. while writing blocks of data) and the startup and tear-down costs of a parallel section. These costs are likely platform dependent.

Compression algorithms in fst are fast on a single thread. That means that performance will be dominated by IO throughput when using only a few cores in parallel. Using more cores will degrade performance due to threading overhead. The ideal (maximum) number of threads to use should be determined per column type.