Fig. 2
From: SparkBLAST: scalable BLAST processing using in-memory operations
The workflow implemented by SparkBLAST: during each of the three stages, parallel tasks (represented as vertical arrows) are executed in the computing nodes. Pre-processing produce the splits of the query file and copy them to the DFS. The main processing execute local instances of BLAST on local data. Finally, the post processing merges output fragments into a unique output file