The statistics are now generated by two different processes: the I/O daemon and the user process. However, everything is handled in one class, which means that statistics objects are often created but only half populated. We may consider to split the class definition of statitics object to better reflect the underlying implementation.