关键词: 海量数据, Top-K, 循环阈值算法（RRTA）, 顺序读取, 有限内存

Abstract: Top-K query is a widely used operation, which can return the highest score of the tuple in specialized massive databases, according to the given monotone aggregation function. Traditional TA algorithm requires the ability to support random access, NRA algorithm although relaxes the restrictions on the random access, it is found that in massive data context, NRA needs to maintain large quantity of candidate tuples in memory in the increasing phase, it will also use up a lot of memory resources in runtime. Compared with the NRA algorithm, the RRTA（Round-Robin Threshold Algorithm） which proposed in this paper replants the data storage mode, which creats a new table to switch the memory overhead to the cheaper external memory overhead, so just sorted access is also able to do efficient top-k query. Meanwhile, the table has been divided, which makes the algorithm more efficient and smoother even with limited memory, in the case of parallel processing.

Key words: massive data, Top-K, Round-Robin Threshold Algorithm（RRTA）, sorted access, limited memory