Fig 1.
Overall workflow of the MapReduce paradigm.
This workflow shows several map and reduce tasks working in parallel over different input splits.
Fig 2.
Overview of the HSRA workflow for single-end alignment.
This mode executes a map-only job taking advantage of the HSP library to parse the reads directly from HDFS. Native pipes are used for efficient IPC communication between Hadoop and HISAT2.
Fig 3.
Overview of the HSRA workflow for paired-end alignment using the reduce-side join approach.
This approach executes a MapReduce job using the single-end support provided by the HSP library, where a reduce-side join is needed to obtain the paired-end reads. Native pipes are used for efficient IPC communication between Hadoop and HISAT2.
Fig 4.
Overview of the HSRA workflow for paired-end alignment using the map-side join approach.
This approach allows avoiding any data shuffling by executing a map-only job thanks to the specific support for paired-end datasets provided by the HSP library. Native pipes are used for efficient IPC communication between Hadoop and HISAT2.
Table 1.
Input datasets used in the experimental evaluation.
Fig 5.
Experimental results for single-end alignment.
Runtime results obtained by HSRA when varying the number of nodes using the (a) SRR1 and (b) DRR1 datasets.
Fig 6.
Experimental results for paired-end alignment (SRR1 dataset).
Runtime results obtained by HSRA when varying the number of nodes using the (a) reduce-side and (b) map-side join approaches.
Fig 7.
Experimental results for paired-end alignment (DRR1 dataset).
Runtime results obtained by HSRA when varying the number of nodes using the (a) reduce-side and (b) map-side join approaches.
Table 2.
Experimental results for DistMap and HSRA using HISAT2.
Table 3.
Experimental results for DistMap and HSRA (16 nodes) vs HISAT2 (1 node).
Table 4.
Experimental results for HISAT2, DistMap and HSRA (4 nodes).
Table 5.
Mapping statistics for HISAT2 and HSRA (SRR1 dataset).