Automatically Configuring Parallelism for Hybrid Layouts

Abstract

Distributed processing frameworks process data in parallel by dividing it into multiple partitions and each partition is processed in a separate task. The number of tasks is always created based on the total file size. However, this can lead to launch more tasks than needed in the case of hybrid layouts, because they help to read less data for certain operations (i.e., projection, selection). The over-provisioning of tasks may increase the job execution time and induce significant waste of computing resources. The latter due to the fact that each task introduces extra overhead (e.g., initialization, garbage collection, etc.). To allow a more efficient use of resources and reduce the job execution time, we propose a cost-based approach that decides the number of tasks based on the data being read. The proposed cost-model can be utilized in a multi-objective approach to decide both the number of tasks and number of machines for execution.