A centralised SCM stores all of its metadata in a single authoritative (or "master") database that is not replicated (except possibly for backup purposes). CVS (and RCS and SCCS) follows this model, as does Subversion.
Mercurial is a DistributedSCM, which is a more general model that's a proper superset of the centralised SCM model. If you want to change your development pattern as little as possible and switch from CVS or Subversion for reasons of non-brokenness (CVS does not have atomic commits), speed, or future flexibility, Mercurial can be used in a way that's very similar to most centralised SCM systems. Instructions on how to do this can be found at CvsLikePractice.