In this paper, a novel approach known as DistFSM is presented for the FSM on a single graph. The DistFSM operation performed on a cloud computing system is framed on a set of heterogeneous clusters. Each cluster is a set of homogenous nodes. The input graph is converted into a sparse matrix. This matrix is partitioned horizontally into a sequence of non-equivalent chunks. Each chunk size is computed to be appropriate to the available worker resources in one of the clusters. In each cluster, the chunk is partitioned vertically into equivalent tasks. Each task is assigned to one of the worker nodes. The proposed partitioning method defined as the Hori-Vertical partition and aims to accomplish the load balancing among the different nodes in the different clusters. Each node performs its operation individually without any communication with other nodes. The non-equivalent chunks assigned to the different clusters allow them to finish their operation simultaneously. This strategy increases the resource usage by prohibiting or reducing the waiting time of the high-performance clusters. Finally, the results of all clusters are summarized and submitted to a distributed shared memory of the orchestration node to perform the required aggregation operations.
Digital Object Identifier (DOI)
Elshrkawey, M.; E. Refaat, Hosam; and H. Amin, Hanan
"A Novel Distributed Approach for Frequent Subgraphs Mining Across Cloud Computing System (DistFsm),"
Applied Mathematics & Information Sciences: Vol. 14
, Article 14.
Available at: https://dc.naturalspublishing.com/amis/vol14/iss2/14