Source : Free On-Line Dictionary of Computing

demand paging
     
         A kind of {virtual memory} where a {page}
        of memory will be {paged in} if an attempt is made to access
        it and it is not already present in main memory.  This
        normally involves a {memory management unit} which looks up
        the {virtual address} in a {page map} to see if it is paged
        in.  If it is not then the {operating system} will page it in,
        update the page map and restart the failed access.  This
        implies that the processor must be able to recover from and
        restart a failed memory access or must be suspended while some
        other mechanism is used to perform the paging.
     
        Paging in a page may first require some other page to be moved
        from main memory to disk ("paged out") to make room.  If this
        page has not been modified since it was paged in, it can
        simply be reused without writing it back to disk.  This is
        determined from the "modified" or "dirty" flag bit in the
        {page map}.  A {replacement algorithm} or policy is used to
        select the page to be paged out, often this is the {least
        recently used} (LRU) {algorithm}.
     
        {Prepaging} is generally more efficient than demand paging.
     
        (1998-04-24)