On Memory and I/O Efficient Duplication Detection for Multiple Self-clean Data Sources