Syntax options:

    RANDOMIZE | NORANDOMIZE

    The RANDOMIZE clause specifies that the sequence numbers are to
    be returned with a random value in the most significant bytes
    of the BIGINT value. This allows unique values to be generated
    that have a random distribution. When you specify the NORANDOMIZE
    clause, sequence numbers are close in value to others created at
    the same time.

    The advantage of the RANDOMIZE clause is that updates to columns
    of a stored index to which these values are written occur in
    different locations in the index structure and so may improve
    concurrent access for large indexes as leaf nodes in different
    parts of the index can be updated independently. In contrast,
    the sequence numbers generated when you specify the NORANDOMIZE
    clause are likely to be close in numeric value to other sequence
    values generated at the same time. This may cause index updates
    to occur in the same or nearby index nodes, which may lead to
    contention in one part of the sorted index.

    The full range of values in the BIGINT value returned for the
    sequence are used; therefore, the NOMAXVALUE and NOMINVALUE
    clauses must be specified (or defaulted to) for the sequence
    definition. The most significant bits of the BIGINT value
    are set to a randomly generated positive value. A generated
    distinct value is returned in the least significant 32 bits so
    that uniqueness is guaranteed. If you also specify the CYCLE
    clause, then only the least significant 32 bits are cycled. When
    a query is performed on the column RDB$NEXT_SEQUENCE_VALUE in
    the RDB$SEQUENCES table, only the generated value of the least
    significant bits is returned, because the most significant bits
    are not assigned until the NEXTVAL pseudocolumn is referenced.

    If you specify RANDOMIZE, you cannot also specify ORDER,
    MAXVALUE, or MINVALUE.