“…The program needs time for both phases (generation and testing), in which the number of generated candidates is very large (Nguyen et al, 2020), the domain of each candidate to be tested is also very large, and thus the time needed for the two phases is very significant. - The large memory requirements: the number of candidate subgraphs is huge (Nguyen et al, 2020), which the system needs to store and evaluate; and for big datasets, the domain to store each candidate is also large, and thus the mining processes consume a lot of storage space.
To improve the performance of GraMi for FSM, there are many algorithms like (Elseidy et al, 2014; N.‐T. Le et al, 2020; R. Li et al, 2018; Nguyen et al, 2020; Wang et al, 2019) to reduce a large portion of redundant candidates, reduce the running time or memory requirements. In 2016, ScaleMine (Abdelhamid et al, 2016) was proposed as a parallel FSM, and then SSIGRAM (Qiao et al, 2018) was introduced in 2018 which was a new parallel approach using Spark.…”