Speedup factor (in terms of overall wall time) obtained by running ddfacet on multiple cores. The solid line corresponds to fixed affinities (each worker process was pinned to a single CPU core), while the dashed line to no affinities (the OS scheduler was allowed to migrate processes across cores). In the former case, processes 0–7 were pinned to the first physical CPU, and 8–15 to the second CPU. This explains the slightly better performance in the no-affinities regime with ≤8 processes, as the OS scheduler was allowed to make use of the second CPU. The graph also shows significanly worse scaling in the hyperthreaded regime.