SGMiner: A Fast and Scalable GPU-based Frequent Pattern Miner on SSDs

Frequent itemset mining is extensively employed as an essential data technique. Nevertheless, the size grows, applicability of this method decreases owing to relatively poor performance existing methods. Though numerous efficient sequential frequent methods have been developed, that can be achieved clearly limited by fact they exploit only one thread. To overcome these limitations, a number parallel using multi-core central processing units (CPUs), multiple machines or many-core graphic (GPU) proposed. However, are slow in and low scalability, mainly large memory requirements for intermediate data, significant disk I/Os, heavy computation. In study, resolve aforementioned problems, we propose SGMiner, which new, fast, scalable GPU- disk-based on single machine equipped with (GPUs) solid-state drives (SSDs) extracting patterns. It based algorithm similar Apriori neither has nor I/O overheads its exploitation SSDs. Moreover, storing transaction databases, namely bitmap chunks, SSDs, streaming chunks GPU device via main reduced overhead, performing fast support counting GPUs chunks. addition, when exploiting it proposes concept replicating stored SSDs fashion. This could allow almost equal workload distributed evenly across overheads. The experiments conducted demonstrate SGMiner outperforms terms scalability enhanced robustness.